VOLUME 14, ISSUE 12, DECEMBER 2025

Abstract: The rapid evolution of artificial intelligence (AI) has significantly transformed modern higher education, particularly in areas related to personalized learning and adaptive instruction. AI-driven systems enable universities to analyze learner behavior, diagnose strengths and weaknesses, and deliver instructional pathways tailored to individual performance. These technologies—such as intelligent tutoring systems (ITS), predictive learning analytics, and adaptive learning platforms—have led to measurable improvements in engagement, retention, and academic outcomes. This paper explores the role of AI in enhancing learning quality, supporting institutional decision-making, and enabling data-driven pedagogical interventions. It further highlights practical applications, global case studies, implementation challenges, and emerging trends such as generative AI in education. The findings confirm that AI is not a replacement for educators; rather, it strengthens their capabilities by automating routine tasks and enriching the learning experience through personalized, flexible, and evidence-based approaches.

Keywords: Artificial Intelligences, Education, Intelligent Tutoring Systems.

Abstract: The proposed system outlines a novel camouflage-enabled mobile robot designed for various roles that require secure and rapid audio, video, navigation, data transmission and the capabilities to camouflage with any background to aid in remote monitoring. The robot is equipped with a self-sustaining power management and obstacle detection system. The robot is designed for military applications and security operations. These military-grade robots are clandestinely deployed and outfitted with a suite of technologies including cameras, sensors and camouflage. The design addresses the challenges of converting operations by integrating state-of-the-art Visible Light Communication Technology for secure and rapid communication. The system’s performance is analysed using statistical methods and Little’s theorem, providing insights into operational efficiency and transmission reliability. The combination of camouflage technology, wireless control, and intelligent surveillance makes this system a valuable asset for defence, security, and research applications.

Keywords: Visible light communication, Camouflage technology Little’s theorem, Military-grade robots, Intelligent surveillance

Abstract: Rapid identification of emergency vehicles is crucial for enabling timely intervention and reducing the risk of accidents in urban traffic environments. Accidents are inevitable in our daily life. To minimize the occurrence of accidents, implementing efficient and well-managed traffic systems is essential. A surveillance system called smart detection of emergency vehicles can identify emergency vehicles that are stuck in traffic. This system supports smarter traffic management in response to the growing number of vehicles on the road in recent years, which has led to increasing congestion. In this paper, we present a prototype of a traffic control system designed to manage signal lights at a junction. When an emergency vehicle approaches, the system temporarily overrides the normal signal cycle and prioritizes its movement by indicating its entry. This ensures that the emergency vehicle can pass through the junction in the shortest possible time. When an emergency vehicle enters, the system will stop the present status of work temporarily and will indicate the entry of the emergency vehicle. So that it can pass through the junctions in a lowest possible time. To achieve this we have utilised a Camera module which helps in identifying or capturing the real time images in the traffic, this camera module is paired with Raspberry Pi 4 as the main controller, the software is implemented in Python, utilising the YOLO framework, and the final implementation is done through a led module which helps in regulating the traffic signals. Since it the proto type the desired implementation is done for only one of the line of the traffic junction.

Keywords: Machine Learning, Image Processing, Segmentation, Early Detection, Artificial Intelligence.

Abstract: Smart hospitals represent the next critical stage in the digital transformation of global healthcare infrastructure. With rapid advancements in the Internet of Things (IoT), Artificial Intelligence (AI), cloud computing, and pre- dictive analytics, modern hospitals are evolving from reactive and manual systems into integrated, automated, and intelligence-driven environments. This paper presents a deeply expanded, hyper-detailed, IEEE-aligned framework for implementing smart hospital ecosystems. It includes an extensively enriched discussion of IoT device integration, multi-layer architectures, communication technologies, edge intelligence, cloud–AI pipelines, clinical decision support systems, and end-to-end workflow automation. The paper expands each conceptual layer with comprehensive technical explanations, extended clinical use-cases, and detailed archi- tectural principles. Additional tables examine communication standards, operational risks, device metrics, comparative models, and implementation trade-offs. This extended version preserves the original content while broadening it substantially, making the work suitable for publication, large-scale technical reports, and academic implementation studies.

Keywords: Smart Hospitals, IoT, Edge Computing, Cloud AI, Healthcare Systems, Predictive Analytics, Clinical Monitoring, Interoperability.

Abstract: The Neo Sentinel Rover is a compact security robot developed using a Raspberry Pi 4 and a set of basic sensors. It continuously monitors its surroundings using an ultrasonic sensor, fire sensor, sound sensor, and metal detector. When any unusual activity is detected, the rover activates a buzzer and immediately captures images or streams live video for remote monitoring. It also sends alert messages to notify the user in real time. This system offers a low-cost and reliable solution for security patrol, border monitoring, and remote surveillance, reducing the need for human presence in risky areas. Its simple design makes it easy to maintain, while its modular structure allows future upgrades and additional intelligence features.

Keywords: Raspberry Pi, Internet of Things, Security Rover, Video Streaming, Ultrasonic Sensor, Remote Surveillance.

Abstract: Diabetic retinopathy (DR) is a leading cause of vision impairment globally, emphasizing the urgent need for early and accurate detection methods. Recent advancements in deep learning (DL) have demonstrated significant potential in automating DR diagnosis from retinal fundus images, thereby aiding clinicians in timely intervention. Nevertheless, the opacity of DL models remains a barrier to their widespread clinical adoption, necessitating transparent and explainable solutions. This paper proposes an integrated framework that combines state of the art deep learning architectures with explainable artificial intelligence (XAI) techniques, specifically Grad-CAM, to improve the interpretability of the diagnosis process. The methodology involves training multiple DL models, including a novel customized convolutional neural network (CNN), on high-resolution fundus image datasets, complemented by extensive data augmentation and preprocessing strategies to address class imbalance and image variability. The incorporation of XAI enables visualization of model decisions, fostering trust and facilitating clinical validation. Experimental results demonstrate that the proposed approach achieves high classification accuracy, superior early-stage detection capabilities, and meaningful interpretability insights, potentially enhancing clinical decision support systems for diabetic retinopathy.

Keywords: Diabetic Retinopathy, Retinal Fundus Images, Grad-CAM, Customized Convolutional Neural Network.

Abstract: The dairy industry faces persistent challenges in ensuring milk quality, transparency in transactions, and operational efficiency. This project proposes an IoT-based system that automates milk collection, quality assessment, and billing processes to address these issues. The system integrates sensors to measure key milk parameters such as fat content, SNF (SolidsNot-Fat), temperature, and volume in real-time. These readings are transmitted to a central microcontroller, which processes the data and calculates the payment based on predefined quality and quantity metrics. The results are displayed instantly and stored in a cloud database for transparency and traceability.

Keywords: IoT, Milk quality management, Billing automation, Auto-filling, Dairy industry.

Abstract: The Text Based Adventure Game project aims to modernize the classic text-adventure genre by integrating AI-powered storytelling, speech recognition, face and eye detection, and multilingual support. Unlike traditional systems where players type commands, this system allows natural voice input to navigate the story. The game intelligently narrates scenes, responds dynamically to player actions, and adapts its difficulty level based on the player's facial expressions. This document outlines the system's design, including its use of the Gemini API for narration and MediaPipe for eye-tracking controls.

Keywords: Artificial Intelligence, Text-Based Games, Speech Recognition, Emotion Detection, Gemini API, Human-Computer Interaction.

Abstract: Computer Vision Syndrome (CVS) has emerged as a critical health concern in the digital era due to prolonged exposure to screens, resulting in eye strain, dryness, and blurred vision. Traditional diagnostic methods rely on clinical examination, which can be invasive, time-consuming, and inaccessible for frequent monitoring. This paper introduces Eyelume, a real-time CVS detection and monitoring system that leverages Vision Transformers (ViT) for accurate pupil segmentation and pupillometry analysis. Unlike Convolutional Neural Networks (CNNs), ViTs capture global dependencies within visual data, offering robustness against low-quality and noisy eye images. The system enables users to upload eye images via a web interface, validates input quality, and computes pupil size variations to identify abnormal responses linked to CVS. Experimental evaluations demonstrate a segmentation accuracy of 99.6%, proving Eyelume’s potential as a non-invasive, accessible, and effective tool for early CVS detection and digital eye health monitoring [1][2].

Keywords: Computer Vision Syndrome, Pupil Segmentation, Vision Transformers, Pupillometry, Digital Eye Health.

Abstract: InclusiMeet is an accessibility-first, full-stack video conferencing system that combines Next.js 14 (App Router, server components) with TypeScript, Tailwind CSS, and shadcn/ui for a responsive, maintainable UI, Clerk for secure multi-provider authentication and protected routes, and a video SDK for low-latency real-time A/V, recordings, device controls, screen sharing, layouts, reactions, and participant management at scale. The architecture uses route groups, dynamic segments for meeting IDs, and layered layouts to separate meeting rooms from the dashboard shell, applying client components only where interactivity is required. InclusiMeet implements instant meetings, schedulable sessions with shareable links, personal rooms, and a recordings hub, while enforcing role-based permissions and session policies. The result is an enterprise-ready, developer-efficient platform that delivers reliable media performance, strong security and privacy, and inclusive UX across desktop and mobile.

Abstract: The project involves the development of two autonomous robots designed to play a soccer game on a physical field. The robots are equipped with cameras for real-time ball and goal post detection, and use pushers to move the ball towards the goal. The game environment consists of a soccer field with a moving goalkeeper, which the robots must avoid while attempting to score goals. Image processing algorithms and real-time decision-making strategies guide the robots’ actions. The system also integrates a web-based control and monitoring dashboard that enables users to view the live camera feed, control the robots manually, and monitor gameprogress in real-time. To play soccer without human control. The robot can detect the ball, recognize goals, avoid obstacles, and make smart decisions during the game. Using sensors and a camera, it sees the field, finds the best path, and moves towards the ball or goal. AI helps the robot learn, plan strategies, and react quickly to changes during play. The robot is designed for smooth movement, teamwork (in multi-robot setups), and real-time performance. This project shows how robotics and AI can work together in fun and competitive environments like robot soccer.

Keywords: Autonomous robotics, computer vision, AI-driven algorithms, deep reinforcement learning, swarm intelligence, multi-agent collaboration.

Abstract: Detection of Brain Tumor Through Retina proposes a non-invasive method for early brain tumour detection using retinal imaging. Since brain abnormalities often affect the retina, featureslike papilledema and optic atrophy are analyzed using fundus and OCT images. A deep learning model is trained to detect these signs, with key features like disc swelling and nerve fibre thinning extracted automatically. A web- based interface enables clinicians to upload retinal images and receive real-time diagnostic predictions. This approach offers a cost-effective, accessible alternative to traditional brain imaging, aiding in early diagnosis and intervention. The system leverages convolutional neural networks (CNNs) to achieve high accuracy in identifying visual biomarkers. It incorporates Grad-CAM heatmaps to enhance model interpretability for clinicians. Retinal datasets are pre-processed for quality enhancement and standardized input. The model undergoes extensive validation using labeled clinical datasets. Predictions are supplemented with confidence scores to support clinical decision-making. This innovative framework bridges ophthalmology and neurology, transforming retinal scans into a powerful diagnostic tool.

Keywords: Brain Tumor Detection; Retinal Imaging; Papilledema; Deep Learning; Fundus Image Analysis; Non-Invasive Screening; Convolutional Neural Networks

Abstract: This paper presents Voice-Based 2D AI Character Mock Interview Assistant, an intelligent and animated system for automated interview practice. The system integrates a resume parser, Applicant Tracking System (ATS) scoring, Large Language Model (LLM)-driven question generation, real-time audio dialogue, and a 2D avatar with lip-sync animation, delivering personalized mock interviews with instant feedback. The platform simulates a realistic interview environment where users interact verbally with an AI interviewer capable of generating dynamic, domain-specific questions based on their resumes. Speech recognition and text-to-speech modules enable natural voice communication, while real-time lip synchronization enhances engagement and immersion. The system also includes anti-cheating mechanisms through webcam monitoring to ensure authenticity during sessions. Experimental evaluation demonstrates that the assistant improves users’ communication confidence and response quality through interactive feedback and resume optimization. This project highlights the potential of AI-driven conversational avatars in education, recruitment training, and career development tools.

Keywords: AI-driven mock interview, 2D animated avatar, Applicant Tracking System (ATS), speech recognition, text-to-speech synthesis, natural language processing (NLP), real-time feedback, large language models (LLMs), virtual interview assistant.

Abstract: The prediction of stock prices is a complex task due to the influence of numerous volatile and non-linear factors. While traditional machine learning algorithms like Extreme Gradient Boosting (XGBoost) are powerful tools for structured data, they often struggle to inherently capture the temporal dependencies in financial time series. This study investigates the application of a Long Short-Term Memory (LSTM) network, a deep learning architecture designed for sequential data, for predicting stock prices. We conduct a comparative analysis, benchmarking the LSTM's performance against a strong XGBoost model on historical data of [e.g., Apple Inc. (AAPL)]. The methodology involves meticulous data preprocessing, feature engineering for XGBoost, and sequence modeling for LSTM. Results demonstrate that the LSTM model significantly outperforms the XGBoost benchmark, achieving a lower Mean Absolute Percentage Error (MAPE) of [LSTM MAPE]% compared to [XGBoost MAPE]%. This finding underscores the strength of models like LSTM and XGBoost in automatically learning temporal patterns and long-term dependencies without the need for extensive manual feature engineering.

Keywords: Stock Price Prediction, LSTM, XGBoost, Time Series Forecasting, Deep Learning, Machine Learning, Comparative Analysis.

Abstract: This project proposes an Artificial Intelligence system for the early diagnosis and classification of bone cancer using deep learning methods, specifically Convolutional Neural Networks (CNN). The system processes medical imaging inputs like X-rays, MRIs, and CT scans. The methodology involves a pipeline of image preprocessing, tumor segmentation, feature extraction, and finally, benign or malignant classification. The solution achieves high performance, demonstrating its potential to assist radiologists and healthcare professionals by providing fast and reliable results. The system also incorporates cloud storage and a web-based interface, making it a scalable and efficient tool for telemedicine applications.

Keywords: Bone Cancer Detection, Convolutional Neural Network (CNN), Deep Learning, Medical Image Segmentation, Tumor Classification.

Abstract: Generative Artificial Intelligence (Gen AI), like other business technologies, has rapidly expanded worldwide. It has transformed organizational tasks and management, emphasizing the need to explore its effects on productivity and employment dynamics. When used as a data processing tool, Gen AI integrates various tasks with professional activities. Consequently, its adoption impacts employees' experience, workload, autonomy, scope of work, skill deployment, and other factors. We have studied the impact of systematically adopting Gen AI on performance metrics and employee well-being, identifying indicators such as productivity gains and challenges in workplace transformation. Using a multi-dimensional, high-volume dataset of 100,000 companies across 14 countries and various sectors, we find evidence of an average increase of approximately 18.47% in productivity, with significant variations across industries such as Defense and Retail. Conflicting reactions and feelings among employees were prevalent alongside productivity gains and concerns about employment preservation. The results showed no statistically significant relationship between training hours and productivity change, emphasizing the importance of strategic application. We employed a hybrid methodological framework combining quantitative and qualitative analysis techniques. Applying descriptive statistics, sentiment analysis, and clustering techniques to examine metrics such as productivity change, employee impact, training hours, and thematic evidence. This study aims to measure the pragmatic justifications of Generative Artificial Intelligence. Further, the study aims to examine cross-sectoral and regional heterogeneity testing emotional responses of the employees via sentiment analysis Reviewing the existing empirical evidence highlights the importance of developing an operational understanding, fostering problem-solving skills, and promoting collaboration with employees, as well as sharing benefits with both employers and workers. This approach enhances the advantages of implementation while carefully addressing concerns related to human capital. This study, through an in-depth analysis, makes a meaningful contribution to the existing literature on Artificial Intelligence-driven organizational adjustment. It also offers specific recommendations to policymakers and industry experts navigating the complexities of technological globalization.

Keywords: Generative AI, Enterprise Productivity, Workforce Adaptation, Sentiment Analysis, Machine Learning

Abstract: Deepfake growth at an accelerating rate presents major threats to security, privacy, and digital media authenticity. Standard approaches to deepfake detection using convolutional neural networks (CNNs) are very good at detecting spatial artifacts but not at detecting temporal inconsistencies between video frames. To overcome this issue, we introduce a hybrid CNN-LSTM deepfake detection model that leverages the best of CNNs for spatial feature extraction with long short-term memory (LSTM) networks for learning temporal sequences. Our model was trained and tested on the celeb-df dataset, which is one of the hardest benchmarks for deepfake forensics. Experimental outcomes prove that the hybrid model outperforms single CNN and LSTM baselines in terms of better accuracy, precision, recall, and F1-score. Results prove the efficacy of combining spatial and temporal modelling for deepfake detection and emphasize the promise of the approach for multimedia forensics and security in real-world applications.

Keywords: Deepfake detection, Hybrid Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM), Celeb-df, Multimedia forensics

Abstract: The development of human-computer interaction (HCI) is constantly pushing the limits of digital art creation. Despite this, the art of creating digital art continues to rely heavily on conventional input devices such as the stylus and the mouse, which may be a deterrent to many. This paper presents "Futuristic Digital Art," a revolutionary AI-powered painting software that turns an ordinary webcam into a smart and emotive creative artistic device. The system uses real-time vision-based hand tracking to provide natural gesture controls for brush and color selection, drawing, and other operations on the canvas. The core of its innovation is a sophisticated auto-correction engine capable of detecting and perfecting hand-drawn geometric shapes and letters with high accuracy. By utilizing the MediaPipe framework for reliable hand landmark detection and an algorithmic solution for gesture understanding and shape identification, the system provides an intuitive, hardware-independent solution. This work shows an important advancement towards democratizing digital art and interaction by combining the expressiveness of natural human motion with the accuracy of smart computational guidance.

Keywords: AI Painting, Gesture Recognition, Human-Computer Interaction, Digital Art, MediaPipe, Shape Correction, Computer Vision.

Abstract: Visual Question Answering (VQA) is a complex multimodal task that requires instant understanding of visual content and natural language queries, yet traditional models often struggle to construct a complete semantic representation of the same. Although conventional VQA systems rely on deep visual feature extraction and linguistic encoders for the question, they commonly fail to capture global context, exact object interactions, and long-range dependencies. A major limitation across early VQA models is the presence of strong language bias, where the system predicts answers based on frequently occurring question-answer patterns rather than genuine visual grounding. To address these issues, recent research has introduced image captioning as a complementary semantic modality capable of enriching scene understanding. Captions provide descriptive information about object attributes, relationships, and contextual cues that may be missing or underrepresented in raw visual features, and integrating them through attention mechanisms such as Attention Aware modules or Question-Guided Parallel Attention allows models to filter irrelevant tokens and retain meaningful semantics. This fused representation creates a more robust and contextually aligned multimodal embedding that strengthens reasoning across diverse question types. Experimental results on benchmark datasets show that caption-enhanced approaches offer consistent improvements in accuracy and interpretability, although they remain dependent on caption quality and introduce additional computational complexity. Nonetheless, the integration of caption-generated semantics represents a promising direction toward developing more context-aware and visually grounded VQA systems capable of more reliable and human-like reasoning.

Keywords: Visual Question Answering, Attention Aware, Question-Guided Parallel Attention, Image Captioning, Deep Learning, VQA v1,VQA v2.

Abstract: The increasing reliance on mobile technologies and rapid adoption of IoT has opened new avenues for modernizing real-time monitoring and management of transportation systems. This paper addresses the long-standing challenges of uncertainty, safety, and communication among the users of transportation in educational institutions. The proposed framework is an IoT system integrating GPS devices mounted on college buses with tracking server and mobile application, enabling seamless communication between devices and end-users. Developed on Android™ platform in Kotlin and enhanced through Google Map APIs, the mobile application collects location data from the GPS device and communicates it to the application on request, enabling students, staff, parents, administrators, and emergency personnel to monitor bus movement in real time. Key features include live location visualization, estimated time of arrival, bus and driver information, and system security through user authentication and authorization. System evaluation demonstrates accuracy in tracking, reliability of updates, and usability across diverse network conditions. By combining open-source IoT tracking platforms with modern mobile technologies, the system contributes to the broader domain of smart campus solutions, providing a scalable and practical approach to intelligent transportation.

Keywords: Google Map, GPS, IoT, Kotlin, Mobile Application, Tracking, Transportation

Abstract: Heart attack prediction remains a critical component of preventive cardiology, requiring highly accurate and interpretable machine learning (ML) frameworks capable of identifying high-risk individuals before the onset of acute myocardial events. Traditional approaches often suffer from limited diagnostic precision due to noisy clinical attributes, heterogeneous patient data, and the absence of systematic feature-engineering strategies. Recent advancements in ensemble machine learning have demonstrated significant improvements in risk-stratification performance by combining multiple weak or strong learners and extracting the most influential clinical predictors. This study reviews and synthesizes recent developments in heart attack prediction models, focusing on ensemble-based architectures, feature-selection techniques, and hybrid frameworks that integrate clinical, demographic, and biochemical features. Special attention is given to the base methodology, which utilizes Random Forest (RF), stacking, and SelectKBest feature engineering to achieve superior precision, recall, and F1-score compared to contemporary works. While numerous ML models exhibit strong performance in cardiovascular prediction tasks, many report lower accuracy than the ensemble-driven framework presented in the base study, primarily due to dataset imbalance, limited feature optimization, and suboptimal model generalization. Through a comparative analysis of eighteen related research publications, this literature survey highlights the strengths, limitations, and methodological gaps across current heart attack prediction studies, ultimately reinforcing the effectiveness of ensemble ML coupled with robust feature engineering as a powerful strategy for early heart attack risk assessment.

Keywords: Ensemble Machine Learning, Feature Engineering, SelectKBest, Random Forest, Stacking Classifier, Clinical Risk Factors, Cardiovascular Disease Detection, Predictive Modelling, Medical Data Analysis.

Abstract: Urban infrastructure maintenance is often hindered by inefficient complaint reporting systems, leading to delays in addressing critical public issues such as potholes, garbage accu- mulation, broken streetlights, and drainage problems. CivicFix is a cloud-based digital complaint system designed to simplify and automate the grievance redressal process. The platform allows users to report issues by uploading an image, while Google Maps API fetches the location details automatically. A machine learning model then classifies the complaint into categories such as potholes, garbage, streetlights, or drainage, ensuring that it is routed to the appropriate municipal department for resolution. The system features separate dashboards and logins for both users and department officers, allowing users to track complaint statuses and enabling authorities to efficiently manage and resolve issues. Additionally, a public voting mechanism prioritizes urgent complaints, ensuring quicker responses to high-impact problems. By leveraging cloud storage, AI-based classification, and automated routing, CivicFix enhances urban governance, making issue reporting more efficient, transparent, and community-driven. Index Terms: Smart city, urban infrastructure, complaint redressal system, AI-based classification, Google Maps API, cloud computing, Firebase, Web-Based Application, Smart Governance

Abstract: Smart farming in rural areas faces challenges such as animal intrusions, irregular irrigation, and lack of reliable energy. This paper presents Smart Agri Sphere, an IoT-based system integrating intelligent monitoring, intrusion detection, and sustainable power generation. The system uses an ESP32 with ultrasonic, IR, sound, LDR, and soil-moisture sensors to automate field surveillance and irrigation. Intrusions trigger a buzzer, deterrent lights, andi instant Telegram alerts. A Zn/Cu bio-electrochemical cell, powered by agricultural waste juice, provides a renewable off-grid energy source for sensor operation. The integrated design improves crop protection, reduces water usage, and supports sustainable farming. Results indicate accurate detection, efficient irrigation control, and stable micro-energy generation.

Keywords: Smart farming, IoT, intrusion detection, bio-electrochemical cell, ESP32, sustainable agriculture.

Abstract: This project presents an automated plant disease detection system developed using Convolutional Neural Networks (CNNs) to support early and accurate diagnosis of crop diseases. Plant diseases significantly impact global agricultural productivity, and traditional manual inspection methods are often slow, inconsistent, and dependent on expert knowledge. To address these challenges, the proposed system leverages deep learning to classify diseases from leaf images with improved precision and reliability. A large dataset consisting of over 87,000 healthy and diseased leaf images across 38 classes was preprocessed and used to train a custom CNN model. The model effectively extracts spatial features from input images and achieves high performance, recording approximately 99% training accuracy and 97% validation accuracy. The solution is deployed as an interactive web application built with Streamlit, enabling users—particularly farmers and agronomists—to upload leaf images and receive real-time disease predictions. By offering a fast, affordable, and scalable diagnostic tool, this work contributes to smarter agricultural practices, timely disease management, reduced dependency on expert intervention, and overall enhancement of crop health monitoring. The study also highlights the potential of CNN-based systems to transform traditional plant disease diagnosis through efficient, user-friendly, and technology-driven approaches.

Keywords: Plant Disease Detection, Convolutional Neural Networks (CNNs), Deep Learning, Image Processing, Machine Learning, Feature Extraction, Automated Diagnosis, Agriculture Technology, Leaf Image Classification, Training and Validation, Dataset Preparation, Image Preprocessing, Transfer Learning, Streamlit Web Application, TensorFlow/Keras, Real-Time Prediction, Mobile/Field Deployment, Accuracy and Performance Metrics, Sustainable Agriculture, Precision Agriculture, Early Disease Detection, Computer Vision, Data Augmentation, Plant Health Monitoring, Model Evaluation, Classification Models, Disease Recognition System, Web-Based Interface, Model Optimization, Field Images, PlantVillage Dataset, Hyperspectral Imaging, Few-Shot Learning (FSL), Generative Adversarial Networks (GANs), Image Segmentation, Object Detection, Decision Support Systems (DSS).

Abstract: The tree-climbing robot for spraying pesticides automates the spraying process and reduces the need for manual, risky work. It safely climbs trees and applies pesticides with better accuracy, minimizing chemical exposure to farmers. The system ensures targeted spraying to reduce chemical wastage and environmental pollution. It can also carry sensors to detect pest-affected areas for focused spraying. Overall, the robot improves safety, saves time, and supports efficient and sustainable farming.

Keywords: Robot, pesticide spraying, automation, precision agriculture, sensor- based monitoring, sustainable farming, crop protection

Abstract: Illegal poaching of valuable trees such as Red Sandalwood (Pterocarpus santolinas), Teak (Tectona grandis), and other endangered species has emerged as a major ecological, economic, and conservation concern across the world. These trees hold immense value due to their aromatic properties, medicinal applications, and use in high-quality furniture and construction industries. However, rampant illegal logging and smuggling activities have led to rapid depletion of forest resources, biodiversity loss, and significant environmental imbalance. To address these critical challenges, this project introduces an innovative IoT-Based Anti-Poaching Alert System for Valuable Trees, designed to detect and prevent unauthorized cutting or tampering activities in real-time.

Keywords: The IoT-based Anti-Poaching Alert System for Valuable Trees uses smart sensors, real-time monitoring, wireless communication, GPS tracking, and cloud-based alerts to detect illegal tree cutting. The system enhances forest protection, ensures quick response, prevents deforestation, and helps authorities safeguard valuable trees through automated alert notifications.

Abstract: The rapid advancement of deep learning and generative models has led to the proliferation of highly realistic synthetic media, commonly known as deepfakes. These manipulated images and videos pose significant threats to privacy, security, and information integrity. Detecting deepfakes has thus become a critical area of research. This study explores the role of facial feature analysis in deepfake detection, focusing on the subtle inconsistencies and artifacts that distinguish authentic faces from manipulated ones. The integration of machine learning and computer vision techniques allows for the identification of minute discrepancies that are often imperceptible to the human eye. The public also believes in deepfakes, and in these situations, individuals are unable to distinguish between genuine and fake. The purpose of this research is to determine which is right and which is not. The Facial Feature Analysis and Miniature Pattern Dissimilarity Verification model (FFA-MPDV), which combines meso4 for lightweight forgery detection with a capsule network to improve special feature retention, is part of the suggested model in this study. Unlike traditional deepfake detection methods, which often struggle with subtle image modifications, the proposed FFA. This unique combination significantly improves detection performance, achieving an impressive 97.3% accuracy, setting it apart from current state of-the-art techniques and making it possible to identify which photographs are real and which are fraudulent in a matter of seconds.

Keywords: Deepfake Detection, Facial Feature Analysis, Generative Models, Capsule Networks, Spatial Attention Mechanism, Multi-Scale Feature Extraction, Forgery Detection, Deep Learning, Computer Vision, Security and Privacy.

Abstract: Interior design planning traditionally requires professional expertise, expensive software tools, and significant time investment, making it inaccessible for many homeowners and small businesses. With the rapid advancement of artificial intelligence, there is an opportunity to automate interior visualization and design generation in a cost-effective and user-friendly manner. This paper presents an AI Powered Virtual Interior Designer that enables users to upload images of empty rooms and generate realistic, fully furnished interior designs based on selected preferences such as room type, design theme and colour schemes. The proposed system integrates modern web technologies with AI-driven image generation. A full-stack architecture is employed using a React-based frontend for user interaction, a Node.js and Express backend for processing requests, and AI services for design generation and visualization. The system supports multiple room categories and design styles, providing instant visual feedback to users without requiring manual 3D modeling skills. Advanced image processing and generative AI models enhance realism while maintaining fast response times. Experimental evaluation demonstrates that the system produces visually coherent and context-aware interior designs that closely align with user preferences. The application offers an intuitive dashboard for design preview, customization, and export, making it suitable for homeowners, interior designers, real estate professionals, and students. This work highlights the practical applicability of artificial intelligence in creative domains and lays the foundation for future extensions such as real-time 3D visualization, 360degree view and voice assistance.

Keywords: AI-Powered Interior Design, Virtual Interior Designer, Artificial Intelligence, Generative AI, 3D Visualization, 360-Degree Room View, Voice-Based AI Assistant, Smart Interior Planning, Web-Based Design System, Human–Computer Interaction, Personalized Design Automation.

Abstract: NGO Connect – Bridging Volunteers with Impact addresses the challenges faced by nonprofits and volunteers in building meaningful, sustainable collaborations. Nonprofits often encounter fragmented systems, inefficient volunteer management, and a lack of intelligent tools to align opportunities with volunteer skills. Meanwhile, volunteers struggle to find personalized engagements that match their abilities, leading to reduced impact and underutilization of resources. NGO Connect is an AI-based platform that brings NGOs, volunteers, donors, and administrators together in one place. It uses intelligent matching to connect volunteers with suitable activities and helps NGOs manage events, resources, and contributions more efficiently. The platform also provides dashboards, analytics, and recognition features to improve transparency and long-term engagement

Keywords: Volunteer Matching, AI Platform, Events and Resources Management, Impact tracking, Dashboards.

Abstract: Image classification has always been one of the most used and highly researched application of machine learning, while it is very easy for us to easily understand and classify the things we see every time. But, for machines to understand and categorize the same things with near human level accuracy, it requires training on large number of images of those objects and lots of internal calculations. Handwritten digits classification is one part of the whole spectrum of the types of images and their categorizations machines are made to do. This research aims to compare the accuracy of a machine learning algorithm i.e. Support Vector Machine (SVM) with that of a deep learning algorithm Convolutional Neural Networks in handwritten digits classification. In this research, we fed in same dataset containing numerous labelled images of handwritten digits to both Support Vector Machine (SVM) and Convulational Neural Network (with 3 CNN layers). The outcome shows that the cnn algorithm outperforms the svm.

Keywords: Image classification, Machine learning, Support Vector Machine (SVM), Convolutional Neural Network (CNN), Handwritten digits classification, Deep learning, Accuracy comparison, Pattern recognition, MNIST dataset, Image recognition.

Abstract: Artificial Intelligence (AI) is revolutionizing modern technology by introducing intelligent systems capable of learning, reasoning, and decision-making without constant human intervention. This paper explores the impact of AI development on various technological sectors including automation, data analysis, healthcare, transportation, and cybersecurity. The integration of AI technologies enhances efficiency, accuracy, and adaptability while transforming industries through machine learning, natural language processing, and computer vision. From smart assistants and autonomous vehicles to predictive analytics and intelligent robotics, AI enables machines to perform complex cognitive tasks. This paper reviews the current advancements, implementation challenges, and future scope of AI development while addressing concerns such as ethical implications, data privacy, and job displacement. The findings demonstrate that AI is not just a technological innovation but a driving force shaping the future of global progress and digital transformation.

Keywords: Artificial Intelligence, Machine Learning, Automation, Deep Learning, Modern Technology

Abstract: The proliferation of online educational platforms and the demand for increased institutional efficiency necessitate a departure from traditional, fallible attendance systems. Conventional methods, such as manual roll calls and single-mode biometric verification, are inherently susceptible to proxy attendance, consume valuable instructional time, and lack the capacity for proactive academic analysis. This paper presents TRACQUE, an innovative AI-Based Multimodal Attendance System designed to address these critical inefficiencies. TRACQUE integrates robust verification techniques: Face Recognition using LBPH and OpenCV/CV2, Embedded Fingerprint Authentication using an R307S sensor managed by an ESP32 microcontroller, and Barcode Scanning using the ZXing library. This combination creates a highly secure and proxy-proof attendance logging system. The system leverages Machine Learning models, specifically Linear Regression for continuous performance prediction and a Decision Tree Classifier for identifying students at risk of academic underperformance. The architecture ensures real-time data processing and visualization through an intuitive web dashboard built with Python and Flask. Experimental validation reports high accuracy rates, 97.6% for facial recognition and 98.1% for fingerprint accuracy, coupled with a model inference time of approximately 45 ms per face image. By correlating secure attendance logs with internal academic metrics, TRACQUE transforms attendance tracking into a proactive academic management tool, enabling timely, data-driven interventions to enhance student success.

Keywords: Multimodal biometric authentication, face recognition, LBPH, fingerprint verification, machine learning, predictive analytics, attendance system, decision tree classifier, linear regression, early warning system.

Abstract: FINNOVA is a contemporary web-based system with the goal to simplify the financial interactions by uniting the management of user accounts, financial information handling, transactions recording, and notifications. Constructed from Next.js, Prisma, and React (Tailwind CSS for UI), the system is developed to give an easy-to-use interface to users to control their financial activities. This document presents FINNOVA's architecture, system elements, data flows, and evaluates its usability and performance. The findings indicate that the platform is scalable, secure under normal usage patterns, and provides far better user experience than conventional financial management tools.

Keywords: Financial management, Web application, Next.js, Prisma, Transaction tracking, User authentication, Dashboard.

Abstract: Smart home automation systems have gained significant popularity in recent years, enhancing comfort, safety, and energy efficiency in modern households. Traditional smart home systems often rely on mobile applications, voice assistants, or remote controls. However, these approaches can sometimes be inconvenient or inaccessible for certain users. In this project, we propose a gesture based smart home automation system using the Mediapipe library integrated with Raspberry Pi 3B+. A camera captures hand gestures, which are processed in real-time using Python and Mediapipe. Based on the recognized gestures, three electrical appliances are controlled via a 3- channel relay board connected to the Raspberry Pi.

Keywords: Feature fusion, Home automation, Deep learning, IOT, Gesture control.

Abstract: Heart disease is one of the most common and serious health problems in the world today. Predicting it early can help save lives by allowing people to get the right treatment on time. In this study, we compare how well machine learning and deep learning models can predict heart disease using patient data such as age, blood pressure, cholesterol level, and other health factors. Several popular machine learning algorithms like Logistic Regression, Decision Tree, Random Forest, and SVM are tested, along with deep learning models such as ANN, CNN, and RNN. Each model’s performance is measured using accuracy, precision, recall, F1-score, and ROC-AUC. Our findings show that deep learning models generally perform better in terms of accuracy and can capture complex patterns in the data more effectively. However, traditional machine learning models are easier to understand and require less computational power. Overall, this comparison helps highlight the strengths and limitations of both approaches and can guide future work in building better heart disease prediction systems.

Keywords: Heart disease prediction, Machine learning, Deep Learning, SVM, Logistic regression, Decision tree, Random Forest, ANN, CNN, RNN, Outliers, comparative analysis.

Abstract: Indoor navigation remains a major barrier for visually impaired individuals, especially in unfamiliar environments such as public buildings, offices, and hospitals. Most existing systems depend on Bluetooth beacons or RFID tags, which either lack accuracy or require regular maintenance. This work presents a low-cost, room-accurate indoor navigation and assistance system built using Li-Fi based room identification, sensor-based directional estimation, YOLO-powered object recognition, and emergency SOS support. A 3W LED driven through MOSFET circuitry transmits room IDs using Li-Fi at 2000 baud, while a BPW34 photodiode-based receiver decodes the signal and forwards location and orientation data to a Raspberry Pi 5 over HTTP. The Pi processes navigation commands, captures user speech, and generates voice-based guidance. Additional features include real-time obstacle alerting, object identification using YOLOv8s, and a safety button that sends an emergency telegram message with an image and a 5-second audio clip. Experimental evaluation in a four-room demo environment shows reliable Li-Fi detection up to 30 cm in low-light conditions, 90% object recognition accuracy, and an average navigation response delay of 5 seconds. The system demonstrates a practical and scalable solution for autonomous indoor mobility for visually impaired users.

Keywords: Li-Fi, Indoor Navigation, Visually Impaired, Raspberry Pi, YOLOv8, Object Detection, Assistive Technology

Abstract: The rapid advancement of an autonomous surveillance boat designed to enhance port security and management through artificial intelligence vessel monitoring. The project presents a cost-effective solution for maritime vessel tracking and reporting by introducing an autonomous boat equipped boat equipped with Lidar/Ultrasonic Sensor for obstacle detection. The boat efficiently classifies diverse vessels within the port. Complementing this, an Optical Character Recognition (OCR) system identifies hull numbers, cross verifying their presence in the port, and consolidates all patient data on a web server. This holistic approach addresses the challenges of monitoring vessel activities, detecting encroachments, and managing information pertaining to violators within port limits, offering a streamlined and technologically.

Keywords: Lidar / Ultrasonic Sensor, ESP3266, Relay sensor, DHT11 sensor.

Abstract: The Integrated Dam Automation System (IDAS) deals with safety and efficiency challenges in dam management by combining IoT, image processing, and deep learning technologies. The system features crack and leakage detection, water quality monitoring, automated gate control, and emergency alert mechanisms. It uses sensors and an ESP32 microcontroller to enable real-time monitoring and quick responses to environmental changes, aiming to reduce flood risks and structural failures. Mismanagement of dams can lead to catastrophic outcomes due to unforeseen events. Currently, most countries rely on manual systems to monitor and control dams, which are slow and imprecise. To address this issue, a method based on IoT is suggested for monitoring dams and aiding in disaster prevention. Real-time data such as temperature, water level, rainfall, and water flow rates are collected to monitor dam safety. This setup provides efficient alert systems that categorize potential threats into blue (low risk), orange (medium risk), or red (high risk) alerts through a mobile app. With this approach, experts can monitor the situation, respond quickly, and take necessary actions to prevent dangerous consequences. Depending on the situation and requirements, the dam operator can choose to control the gates manually or automatically. This capability simplifies the management of multiple dams and allows for accurate predictions based on the collected data. Drought, which is also a disaster, can be partially managed with dams. The proposed system demonstrates its effectiveness in drought prevention. This work utilizes the Arduino open-source electronic platform. Index Terms: IoT, Dam Automation, Crack Detection, Water Level Monitoring, Deep Learning, ESP32, Image Processing, Turbidity Sensor, pH Sensor, Emergency Alert System, YOLOv5, Smart Infrastructure, Real-Time Monitoring.

Abstract: Air pollution is a major global environmental and health concern, necessitating accurate and timely forecasting of pollutant levels to mitigate adverse effects. Recent advances in machine learning and deep learning enable precise air quality prediction, yet deploying these models in real world resource constrained settings remains challenging. Embedded models augmented with quantization techniques offer an efficient solution by reducing computational costs without significant loss of accuracy. This review synthesizes recent developments in air quality prediction using embedded Machine Learning (ML) and Deep Learning (DL) models with emphasis on 8 bit quantization, hybrid architectures, attention mechanisms, and knowledge distillation. The focal study demonstrates a CNN BiGRU model achieving an R² of 0.99 for PM2.5 on IoT devices, balancing high accuracy with model compression. Through analysis of fourteen contemporary works, this paper highlights multi task learning frameworks, spatial temporal modeling over multiple pollutants, and transformer based approaches as emerging trends. Persistent challenges include extending forecast horizons, improving generalizability, and enhancing real time deployment viability. The review concludes by discussing future directions integrating ensemble methods, Bayesian hyperparameter optimization, and quantum learning potentials toward robust, scalable air quality prediction systems.

Keywords: Air Quality Prediction, Deep Learning, Long Short Term Memory (LSTM), Convolutional Neural Network (CNN), Quantization, Edge Computing, Internet Of Things (IoT), Transformer, Knowledge Distillation, Multi Pollutant Forecasting

Abstract: Edge detection is a foundational operation in computer vision and embedded image processing, enabling systems to identify structural boundaries and salient features within visual data. Real-time implementation of edge detection remains challenging in software platforms due to high computational load and latency. To address this limitation, a hardware-optimized VLSI architecture implementing the Sobel operator is proposed for real-time video edge extraction. The system acquires live frames through a VGA camera, performs grayscale conversion, calculates horizontal and vertical gradients using Sobel convolution masks, and computes edge magnitude before applying thresholding. Implemented on the Artix-7 FPGA using Verilog HDL, the design leverages parallelism, pipelining, on-chip memory optimization, and low-power computation. Experimental evaluation demonstrates significant improvements in throughput, latency, and resource efficiency over traditional software-based methods. The architecture is suitable for embedded vision applications such as surveillance, robotics, and smart IoT cameras. .

Keywords: Sobel, VLSI, FPGA, Edge Detection, Verilog, Real-Time Video Processing.

Abstract: A survey highlights the alarming potential of 700 million people facing displacement due to severe fuel scarcity by 2030. This paper introduces the "Smart Fuel ATM System", a new solution that aims to revolutionize access to safe and secure fuel dispenser. This unique system integrates Radio-Frequency Identification (RFID) technology, Quick Response (QR) codes, Unified Payments Interface (UPI) payments with fuel dispensing gun. The Smart Fuel ATM makes it easy and convenient by using RFID technology for user identification and monitoring fuel dispensing. It enables the user to use cashless transaction; it also ensures compatibility with various financial platforms. This paper provides an elaborated and technical overview of the Smart petrol ATM system, including architecture, dataflow, and synchronization. It includes the above-mentioned technologies highlights their bond in creating an efficient and secure fuel dispensing unit. This dispensing unit has the potential to really improve fuel access, distribution, and quality in both urban and rural areas without any difficulties. By offering safeguarded digital payments and quality control while dispensing the fuel from the system, it discards any kind of critical challenges in safe fuel dispensing, in return contributing to improved fuel dispensing unit for the communities in need. In an overall view of the system, we can make sure there is no error during the payments and during the fuel dispense.

Keywords: IoT, Arduino module, Wi-Fi module, RFID card, Mobile Applications, Real Time Data

Abstract: Handwritten digit recognition is a classic problem in the field of computer vision, and the MNIST dataset is one of the most common benchmarks used to test different machine-learning methods. In this study, we take a closer look at how Convolutional Neural Networks (CNNs) and Support Vector Machines (SVM) perform on this task. SVM, a traditional machine-learning technique, works by treating each image as a long list of pixel values and depends heavily on manually designed features. In contrast, CNNs can automatically learn important visual patterns such as edges, curves, and shapes directly from the raw images. To understand the strengths and weaknesses of each approach, we trained both models on the MNIST dataset and compared their performance using accuracy, precision, recall, and F1-score. Our results show that CNNs consistently outperform SVM, especially when it comes to understanding subtle variations in handwriting. This happens because CNNs are better at capturing the spatial structure of images, something traditional algorithms struggle with. Overall, the study highlights why deep learning models like CNNs have become the preferred choice for image-based tasks, offering a clear advantage over classical machine-learning methods.

Keywords: Image classification, Machine learning, Support Vector Machine (SVM), Convolutional Neural Network (CNN), Handwritten digits classification, Deep learning, Accuracy comparison, Pattern recognition, MNIST dataset, Image recognition.

Abstract: This project presents a multifunctional robotic system designed to support defense and emergency missions by using Raspberry Pi, AI, and DET-YOLO algorithms. The robot integrates a surveillance camera, PIR sensor, metal detector, and voice module for effective detection and communication. It can identify and track aircraft or landmines, providing real time feedback for safer operations. The system uses DC motors for smooth mobility and can navigate rough terrains autonomously. By combining intelligent image detection and terrain sensing, the robot minimizes human risk in combat zones and hazardous areas. Its adaptable design also makes it suitable for civilian uses such as disaster management and search-and rescue operations. The goal of this work is to create a cost-effective, AI-enabled robot that enhances safety and efficiency in high-threat environments.

Keywords: Pir sensor, Laser gun, Camera, DC Motors, H-Bridge, Proximity Sensor, Color sensor, Zigbee, Dispenser kit, Raspberry pi, Ultrasonic sensor, Battery.

Abstract: Deep learning has become a transformative approach for automated plant health monitoring, enabling accurate disease recognition directly from leaf images without relying on manual inspection or expert availability. In modern precision agriculture, both leaf disease classification and severity quantification are essential for identifying early infections and supporting informed intervention strategies. However, developing reliable diagnostic models remains challenging due to environmental variability, heterogeneous field conditions, inconsistent image quality, and the absence of pixel-level severity annotations in standard datasets. This literature-aligned study synthesizes advances in lightweight CNN architectures, classical image-processing pipelines, attention-guided visualization tools, and mobile-centric deployment frameworks for real-time plant disease assessment. Special emphasis is placed on the proposed end-to-end system, which integrates a custom PyTorch-based CNN with an OpenCV-driven severity estimation module and a cross-platform React Native mobile interface. While originally optimized for binary classification, the system directly addresses practical agricultural constraints such as uneven lighting, morphological variations across species, and limited computational resources in field environments. By combining interpretable predictions, severity mapping, and rapid inference via a Flask backend, the approach enhances usability, improves generalization under diverse conditions, and reduces diagnostic dependency on experts. Through comparative analysis with existing methods, this work positions the proposed framework as a promising foundation for future mobile plant-disease diagnostic pipelines integrating accessibility, explainability, and deployment-scale robustness.

Keywords: Plant Disease Detection, Convolutional Neural Networks, Severity Estimation, Mobile Application, Deep Learning, Precision Agriculture.

Abstract: Startup GPS is an AI-powered platform designed to guide early-stage entrepreneurs through the toughest parts of starting a company. Many new startups struggle because of gaps in market research, unclear strategy, or not having the right people on the team. Startup GPS simplifies this early journey by giving founders the tools and clarity they need from day one. At the core of the platform is the Groq LLM, which evaluates business ideas, offers market insights and creates personalized action plans, without need for complicated NLP setups. To help founders build stronger teams, Startup GPS includes a skill-based matching system that connects users with potential members whose strengths complement the project’s needs. The platform also features a Research Finder that pulls relevant academic papers from sources like arXiv, crossref and Semantic Scholar. This allows users to dig deeper into technical knowledge, validate ideas and stay aligned with the latest research trends. By combining idea validation, trend analysis, Team Building and guided strategic planning in one place, Startup GPS removes many of the early barriers founders face. It acts as practical, always available companion for both solo founders and startup teams providing direction, clarity and confidence throughout the entire building process.

Keywords: AI Powered entrepreneurship, Startup Idea Validation, Personalized startup Roadmaps, real time market insights

Abstract: The Solar Sea Weather and Pollution Transmitter Buoy is an important step forward in ocean monitoring technology. It runs on solar power, allowing it to work efficiently and on its own without depending on external energy sources. This not only makes it cost-effective but also environmentally friendly. The project involves creating a solar-powered buoy that can stay afloat and continuously track sea weather conditions and pollution levels in real time. The buoy is equipped with sensors that measure factors such as temperature, humidity, wind speed, and water quality. All the collected data is sent wirelessly to a main control unit, where it can be analyzed and displayed right away. The results show that the buoy works well in delivering fast and accurate environmental information, helping researchers and authorities make better decisions to protect marine ecosystems.

Keywords: STM32, ATMEGA328P Controller, Wireless, IOT and Cloud Computing, Ocean Management, Sensor Suite, Real-Time Data Collection.

Abstract: Nowadays, many educational institutions are dependent on manual attendance and basic ID checks, which are time-consuming, error-prone, and vulnerable to manipulation, leading to inefficiency and security risks. The proposed system - EduFace addresses these issues by using OpenCV and Dlib for realtime facial recognition, enabling automated attendance and secure access control. A Raspberry Pi–controlled motor system manages door access for verified users, while unauthorized attempts are flagged through the web interface. The system also sends WhatsApp alerts to parents via Twilio and provides administrators with a centralized dashboard for real-time monitoring. By reducing manual effort and preventing proxy attendance. EduFace also offers a modern, efficient and secure solution for campus management. In addition, the system currently does not integrate cloud-based scaling, resulting in attendance data and logs being stored locally, which can make management more challenging for very large institutions

Keywords: OpenCV, Dlib, Raspberry PI, Twilio

Abstract: This project introduces an AI-based healthcare robot designed using an Arduino Mega to automate medication delivery and personal care in hospitals and homecare environments. The robot integrates MAX30102 (SPO2) and temperature sensors for real-time health monitoring, an ultrasonic sensor for safe navigation, and a webcam for continuous patient observation and visual feedback. Health data is shown on an LCD display and wirelessly transmitted through Bluetooth/Zigbee for remote monitoring. The APR voice module supports verbal communication, enhancing user interaction. Equipped with motors, a relay-pump mechanism, and smart control, it reduces human workload, promotes contactless healthcare, and provides a reliable, low-cost medical assistant.

Keywords: AI healthcare robot, Arduino Mega, Webcam monitoring, MAX30102 sensor, APR voice module, Wireless communication, Human-robot interaction, Contactless care, Smart healthcare system.

Abstract: Accurate prediction of healthcare insurance costs plays a crucial role in improving cost management, policy design, and healthcare planning. This study investigates the effectiveness of various machine learning (ML) algorithms in forecasting healthcare insurance expenditures and identifies the most suitable model for reliable cost estimation. A publicly available dataset containing demographic and lifestyle-related attributes such as age, sex, body mass index (BMI), number of children, smoking status, and region was utilized. Multiple regression-based ML models, including Linear Regression (LR), Support Vector Regression (SVR), Random Forest Regressor (RFR), XGBoost Regressor (XGBR), LightGBM (LGBM), and Gradient Boosted Regression (GBR), were implemented and compared. The evaluation results demonstrate that the GBR model outperformed other approaches by achieving the lowest mean squared error (MSE = 18,153,562.14) and mean absolute error (MAE = 2,270.97), along with the highest coefficient of determination (R² = 0.87), peak signal-to-noise ratio (PSNR = 22.97), and signal-to-noise ratio (SNR = 9.97). Cross-validation further confirmed its robustness, with the tenth fold achieving an R² of 0.91. To enhance model interpretability, explainable artificial intelligence (XAI) tools such as SHAP and LIME were applied to the final GBR model, revealing that “region” and “smoker” were the most influential factors affecting insurance costs. The findings confirm that GBR, combined with explainable AI techniques, offers a robust, transparent, and reliable solution for predicting healthcare insurance costs. Future work will focus on integrating more advanced explainable frameworks and real-world healthcare datasets to further improve reliability and applicability.

Keywords: Healthcare insurance cost prediction; machine learning; explainable artificial intelligence (XAI); regression models; gradient boosting

Abstract: This research focuses on the development of a Smart Disease Prediction System capable of identifying multiple potential diseases from a single blood sample using machine learning techniques. The system analyzes critical blood biomarkers such as glucose, RBC count, WBC count, platelets, hemoglobin, cholesterol, triglycerides, creatinine, and additional biochemical indicators to classify health conditions with increased reliability. The proposed framework involves systematic data preprocessing, feature extraction, and multi-class classification using trained predictive models, enabling fast medical assessment with minimal manual intervention. A supervised learning model is trained on curated medical datasets and evaluated using accuracy, precision, recall, F1-score, and confusion matrix. The system generates disease risk output along with medically interpretable reasoning based on abnormal parameter deviation, making it useful for early diagnosis. Experimental results demonstrate that the system can detect diseases including diabetes, anemia, heart-related issues, high cholesterol, dengue, kidney disorder, and thyroid variations with promising prediction accuracy. This work aims to support healthcare systems through automation, improving diagnosis speed and reducing dependency on lab evaluation delays.

Keywords: Smart Disease Prediction System; Machine Learning; Blood Parameters; Multiclass Classification; Healthcare Automation; Early Diagnosis; Disease Identification; Biomedical Informatics.

Keywords: Smart Agriculture; Machine Learning; Crop Yield Prediction; Fertilizer Recommendation; Crop Recommendation; Precision Farming; Decision Support System; Artificial Intelligence (AI); Random Forest; Rule-Based System; Data Analytics; Web Application; Flask Framework; MySQL Database; Sustainable Agriculture; Agricultural Informatics; Predictive Modeling; Farmer Advisory System.

Abstract: Applications on Research Papers project The goal of Using AI Agent is to use AI to make it easier to analyze and understand research papers. It uses intelligent agents powered by machine learning and natural language processing to extract, summarize, and classify important information from research documents. Thanks to technology, consumers may use keywords or subjects to find, review, and compare articles with ease. Time and effort can be saved by automating data extraction and processing. The AI agent makes study knowledge more accessible by offering precise summaries and insights. For academics, researchers, and students, this technology facilitates material analysis and decision-making. In general, it encourages appropriate information management and increases research productivity.

Keywords: AI Agents, Automation, Reproducibility, Natural Language Processing (NLP), Scientific Validation, Code Generation, Benchmarking, Knowledge Extraction, Research Transparency, Machine Learning, and Computational Reproducibility.

Abstract: Parkinson’s disease (PD) is a progressive neurological disorder that affects movement control. It is often marked by tremors, stiffness, and slowed motion. Early and precise detection of Parkinson’s disease is vital for effective treatment and management, as it can greatly enhance patients’ quality of life. Recently, machine learning and signal processing techniques have proven promising in identifying Parkinson’s disease using various biomedical signals, including voice recordings, handwriting patterns, and gait analysis. By extracting key features and training classification models, these systems can differentiate between healthy individuals and those with Parkinson’s disease with high accuracy. This study aims to create a reliable detection model that uses data-driven approaches to support medical diagnosis and enable early intervention. The proposed method seeks to improve diagnostic efficiency, minimize human error, and contribute to better healthcare systems.

Keywords: Parkinson’s Disease (PD), Early Detection, Neurological Disorder, Machine Learning, Data Preprocessing, Accuracy and Performance Evaluation.

Abstract: The process of medical diagnostics in developing regions remains fragmented, often requiring patients to manually locate laboratories, compare costs, and retrieve reports across multiple platforms. This paper presents Lab Link, an intelligent full-stack web application designed to simplify and digitize the end-to-end process of medical lab testing. The system allows users to securely create accounts, search for diagnostic tests, and discover nearby laboratories based on location and test availability. Each listed laboratory displays detailed information including test pricing, facilities, and user ratings, enabling data-driven selection. The application integrates Next.js for responsive client- side rendering, Cloudflare Workers for scalable backend computation, and PostgreSQL with Prisma ORM for optimized data management. JWT-based cookie authentication ensures secure and persistent sessions, while Zod validation enforces type-safe operations throughout the system. Test results are stored and retrieved in encrypted PDF format, preventing data loss and ensuring patient privacy.

Abstract: Bioactivity prediction plays a crucial role in contemporary drug discovery, allowing researchers to efficiently pinpoint potential therapeutic candidates while minimizing both experimental costs and development timelines. This paper offers an in-depth exploration of machine learning techniques aimed at forecasting the biological activities of chemical substances against specific biological targets. We evaluate a range of algorithmic methods, including Random Forest, Support Vector Machines, Neural Networks, and Bayesian techniques, assessing their effectiveness across comprehensive datasets. Additionally, the study delves into molecular representation methods, feature engineering tactics, and validation frameworks that are vital for creating reliable bioactivity prediction models. Our findings reveal that machine learning methodologies can deliver remarkable predictive accuracy, with certain algorithms outperforming others based on the characteristics of the dataset. We also examine the integration of extensive databases such as ChEMBL and PubChem, which provide crucial training data for crafting adaptable models. The results underscore both the transformative capabilities and existing challenges of computational bioactivity prediction while offering insights into future research avenues such as explainable AI, transfer learning, and multi-omics integration. This research adds to the accumulating evidence that positions machine learning as an essential resource for expediting pharmaceutical research and lessening reliance on expensive high-throughput screening experiments.

Keywords: Bioactivity prediction, Machine learning, Drug discovery, QSAR, Molecular descriptors, ChEMBL, Deep learning, Random Forest.

Abstract: This paper proposes the design of the system, to detect the ripeness of a fruit using temperature readings and image processing techniques. The temperature module MLX90614 measures the temperature of the fruit and the image processing technique analyses an image of the fruit to determine its color. By combining these two readings, the program can determine whether the fruit is ripe or not. The program collects temperature readings from a Nodemcu device and then analyses an image of the fruit using OpenCV library to get the average color of the fruit. Then, it converts the average color from BGR to RGB and passes it through a function that converts the RGB color to a single value. This single value is then used along with the temperature readings to determine if the fruit is ripe or not. In summary, the project provides a way to automatically detect the ripeness of a fruit by analyzing its temperature and color using image processing techniques. This could potentially be useful for fruit processing and harvesting industries.

Keywords: Image Processing, digital image processing, image detection, VGG16 model, fruits dataset, image classification.

Abstract: FLASHO is a cutting-edge platform that enables on-demand fuel delivery for both corporate and individual customers. Customers can easily order fuel via a mobile application, and it will be delivered promptly to the location of their choice. To enhance the user experience, the platform makes use of secure payment method, real-time notifications, and GPS technology for tracking. FLASHO offers fleet management tools for businesses and drastically reduces waiting times and fuel waste by eliminating the need to visit gas stations. The platform is based on a cloud-based, scalable architecture that respects legal requirements, puts safety first, and encourages eco-friendly behaviour. FLASHO's reliable, effective service is poised to revolutionize the fuel industry.

Keywords: Cloud-based platform, Safe payment method, GPS technology, On-demand fuel delivery, Safety procedures, and Improved user experience.

Abstract: In this paper, a machine learning-based system that helps South Karnataka farmers choose appropriate medicinal crops is presented. It combines a web-based platform that offers real-time recommendations and historical trend storage with a trained Random Forest model. 12,800 samples from 8 classes of medicinal crops and 17 input features, such as soil nutrients, micronutrients, climate, and geographic indicators, are included in the dataset. On a stratified split, the Random Forest classifier’s test accuracy was 58.59%. Water availability, temperature, and pH are important influencing factors. The model was implemented as part of a comprehensive smart farming solution that included a MongoDB database, React frontend, and Node.js backend.

Keywords: Crop Recommendation, Machine Learning, Smart Farming, Medicinal Plants, Random Forest

Abstract: Brain Stroke prediction introduces a robust, multi-modal machine learning system engineered to precisely forecast brain stroke risk by integrating two fundamentally different data sources: conventional, structured clinical data and complex, unstructured CT/MRI neuroimaging. The system is built upon a dual-stream architecture: the gradient boosting algorithm XGBoost is deployed to analyze the patient's record features (e.g., demographics and history), and a deep convolutional network, EfficientNet-B0, is dedicated to extracting visual pathological indicators from the brain scans. A core objective is to ensure system trustworthiness through the application of Explainable AI (XAI), specifically SHAP (Shapley Additive Explanations), which guarantees clarity and interpretability for medical professionals. This scalable solution marks a significant advancement in early stroke detection and enables evidence- based clinical decision support.

Keywords: Explainable AI (XAI), SHAP (Shapley Additive Explanations) , EfficientNet-B0, XGBoost, Convolutional Neural Network (CNN).

Abstract: Pulmonary disorders such as pneumonia remain a major clinical challenge, increasing the need for rapid and dependable diagnostic methods. This study proposes a machine learning–based system that examines both patient-reported symptoms and chest X-ray scans to estimate the probability of pneumonia. An initial risk score is generated using a weighted survey analysis, after which the chest radio-graphs are processed using MobileNetV2 for feature extraction and fed into a Convolutional Neural Network (CNN) for classification. By combining symptom evaluation with automated image interpretation, the system improves diagnostic accuracy and reduces reliance on manual assessment. This integrated approach supports faster screening and enhances the overall efficiency of pulmonary disease detection.

Keywords: Machine Learning, Deep Learning, CNN, MobileNetV2, Chest X-ray.

Abstract: This project presents an AI-powered website/code generation platform built with Next.js, React, and Tailwind CSS supported by Node.js, Axions for API communication. The system allows users to generate fully functional websites and code structures simply by providing natural language prompts. Unlike traditional website builders that require manual coding or drag-and-drop interfaces, this solution leverages artificial intelligence to automate code creation, reducing development time and making web development accessible to both technical and non-technical users. The platform provides a modern, responsive interface where users can enter their requirements, receive auto-generated code in real time, and preview the resulting website. It includes modular components such as a navigation bar, hero section, feature highlights, pricing modules, and a footer, making it a scalable starting point for more complex applications. The system also incorporates a mock API for demonstration, which can later be extended to integrate advanced AI backends (e.g., Open AI, Firebase functions, or custom ML models). From a business perspective, the project has strong potential as a Software-as-a-Service (SaaS) application, offering users a quick and cost effective way to build websites. It can evolve into an AI website builder, a developer productivity tool, or an enterprise solution for rapid dashboard creation. Future enhancements include user authentication, template libraries, drag-and-drop editing, collaboration features, and one-click deployment to hosting platforms.

Keywords: AI-powered code generation ,Website generator, Natural language to code,Next.js, React, Tailwind CSSNode.js, backend, API-based code generation, Axios communication, Automated web development, Real-time code preview, Modular UI components, Navigation bar / Hero section / Pricing module, Mock API integration, SaaS platform, Developer productivity tool, AI website builder, Rapid website creation, Template library, Drag-and-drop editor, Oneclick deployment, User authentication, Scalable architecture, Enterprise dashboard generation, Machine learning integration.

Abstract: Modern software development faces a persistent challenge: translating conceptual requirements into functional backend systems demands specialized technical knowledge that excludes many potential innovators. This paper presents SmartAPIForge, a new platform conceived to overcome this accessibility barrier by intelligent automation of REST API creation. Unlike currently available tools that focus on user interface generation, our work puts greater emphasis on developing the back-end infrastructure with input in natural language. The platform architecture includes a variety of technological advancements in order to realize dependable automation. First, we apply modern language models to interpret what the user needs and then generate appropriate code structures. Second, micro-virtual machines based on Firecracker provide isolated execution environments and protect against security flaws during testing and code development. Third, we keep record of development history during project life cycles and enable team collaboration by integrating with GitHub's version control system.To empirically validate our system, we tested it against 3,079 different API creation tasks derived from real-world development scenarios. The findings show that deployment processes were successful 96% of the time, and generated code compiled successfully in 94% of cases.Furthermore, validation checks against industry standards were passed by 97% of generated OpenAPI specifications. The platform generated everything from the initial prompt to a deployed, documented API within 60 seconds for 95% of test cases. These results show that automated methods can greatly lower the barriers to expertise and time spent on API development. Through careful testing and validation, this work proves that the technology is viable while keeping code quality intact. SmartAPIForge represents a useful step toward making backend development accessible to more users by lowering entry barriers without losing engineering standards.

Keywords: Automated Testing, Code Generation, GitHub Integration, Large Language Model Technologies, Natural Language Processing Techniques

Abstract: Detection for Bone Cancer is a serious medical issue which demands immediate attention and intervention for better patient outcomes. Conventional methods have some serious limitations with regards to accessibility, cost-effectiveness, processing times, and availability on a global scale with respect to specific radiological knowledge and expertise. Raising awareness and promoting research for better AI-based medical technologies with significant societal benefits due to quick intervention and cost-effectiveness, we propose here an innovative dual-architecture AI system named ‘OsteoScan.AI’ combining ResNet18 Convolutional Neural Networks and Google Gemini Generative AI for holistic bone cancer examination. We propose an original seven-layer validation technique effectively rejecting images that are not medical images at all and relate to photography, selfies, and graphics before classification analysis. Utilizing pre-training with ImageNet pre-trained weights on ResNet18 Convolutional Neural Networks, we notice an outstanding accuracy rate of 95.2\% for classification of Bone Cancer from Bone X-Rays as Malignant and Normal classes. It can be effectively implemented as a complete-end stack online platform with React.js GUI implementation, Flask Web-Server implementation for backend with end-to-end processing below 1 second, Medical Image Authentication and Classification package with comprehensive classification and examination scan history, and an ‘AI-Counseling-System’ with an AI-powered chat platform for medical inquiries. EXPLANATION OF EXPERIMENTAL RESULTS clearly authenticates its efficacy and capabilities on strict medical examination criteria within 99.6\% rejection rate on ‘NON-MEDICAL-IMAGE’ classification. Both precision and accuracy with ‘EXPLANATIONS-IN-NATURAL-LANGUAAGE ’attempts to bridge an unsolved gap on critical usage and intervention with Medical AI technologies.

Keywords: Bone Cancer Detection, Deep Learning, ResNet18, Transfer Learning, Medical Image Analysis, Computer Vision, X-Ray Validation, Dual-AI System, Generative AI, Explainable AI, Responsible AI

Abstract: Tuberculosis (TB) remains a serious global health problem, especially in regions with limited access to expert medical care. While chest X-rays are widely used for TB screening, interpreting them accurately can be challenging. This work introduces an automated system that helps detect TB from X-ray images using advanced image processing and artificial intelligence. The system first enhances and isolates the lung areas using the nnU-Net model, then analyzes them with a Swin Transformer to identify signs of infection. Tests on well-known datasets, such as Shenzhen and Montgomery County, showed excellent performance, achieving 95.2% accuracy and a Dice score of 0.94. Overall, this approach offers a reliable and scalable tool that could support faster and more consistent TB diagnosis, particularly in resource-limited healthcare settings.

Keywords: Tuberculosis, nnU-Net, Swin Transformer, Gaussian Filter

Abstract: Smart cities provide a wide range of options for dining, leisure, and essential services, but users often struggle to identify the most suitable choices due to generalized search results and scattered information across multiple applications. This leads to increased decision-making time and reduced convenience. To address this challenge, we developed Explore Hub, a lightweight real-time recommendation platform designed to help users discover restaurants, weekend destinations, and essential locations efficiently. The system integrates Google Places and Geocoding APIs with a React frontend and a Node.js Express backend to deliver accurate place details, interactive map navigation, comparison features, and favorites management using Local Storage. The prototype demonstrates that real-time API-based recommendations, combined with map-based visualization, can enhance user experience and improve decision-making without requiring complex backend infrastructure or machine-learning models.

Keywords: Smart Cities, Recommendation System, Google Places API, React, Node.js, Map Visualization.

Abstract: The catastrophic failure of cellular and internet infrastructure during natural disasters critically undermines emergency response. This paper presents a resilient and self-sustaining communication system designed to operate independently of all traditional networks. The proposed solution is a solar-powered, decentralized mesh network utilizing LoRa (Long Range) technology built on ESP32 microcontrollers. Each node integrates a NEO-6M GPS module for location tracking and hosts a local web server, creating an offline platform accessible via any standard smartphone. The system provides two critical services: a text-based messaging application and a real-time GPS tracking interface, displaying user locations on an offline map. Field tests validate the network's ability to self-form, route messages efficiently, and maintain operation through its integrated solar power system, presenting a practical tool for restoring essential communication in the immediate aftermath of a disaster.

Keywords: LoRa, Mesh Network, Disaster Communication, ESP32, GPS Tracking, Solar Power, Offline Web Interface, Emergency Management

Abstract: Effective food management plays a vital role in maintaining student satisfaction and promoting sustainability in hostel environments. However, many hostels still depend on manual processes such as paper records and spreadsheets, which lead to inefficiencies, food wastage, and poor coordination among stakeholders. MealMap is a digital hostel food management system designed to automate and streamline operations through a unified role-based platform that connects administrators, kitchen staff, and students. The system integrates meal planning, inventory monitoring, and structured feedback mechanisms to enable real-time tracking, automated scheduling, and data-driven decision making. By digitizing hostel food operations, MealMap reduces manual workload, minimizes waste, and enhances transparency and communication among stakeholders. This study presents the design and implementation of MealMap, emphasizing its role in improving efficiency, sustainability, and user satisfaction in hostel food management.

Keywords: Food Management System, Hostel Automation, Meal Planning, Inventory Management, Feedback System.

Abstract: Traditional retail checkout processes are often characterised by inefficiencies, leading to long queues and a decrease in customer satisfaction. This paper proposes "Smart Cart," an end-to-end automated billing system to streamline the in-store shopping experience. The proposed solution utilises a modern web technology stack-MongoDB, Express.js, React, Node- to create a responsive user interface and a robust backend. Key features include real-time barcode scanning through a device camera or hardware integration, dynamic cart management with budget tracking, and a secure wallet-based payment system. To guarantee transaction integrity, the system implements a checkout verification algorithm that cross-checks the software cart data against the hardware sensor data, such as weight and item count. This system also integrates accessibility options and an AI-powered chatbot for nutritional analysis, thus demonstrating a highly scalable and user-centric approach toward the automation of retail.

Keywords: Smart Cart, Automated Billing, Internet of Things (IoT), React.js, Node.js, Generative AI, Web Application, Retail Automation.

Abstract: Endometrial cancer is one of the most common cancers affecting women worldwide. Early detection and accurate grading are crucial for improving survival rates, but traditional diagnostic methods can be invasive and unreliable. This work presents a deep learning approach that combines image preprocessing with Convolutional Neural Networks (CNN) for automated prediction and grading of endometrial cancer from histopathological images. Preprocessing steps, such as converting images to grayscale, filtering out noise, applying thresholds, sharpening images, and segmenting them, help to enhance image quality. The images are used to train CNN models. The study also compares these models with traditional machine learning classifiers like Super Vector Machine (SVM) and K-Nearest Neighbor (KNN). The model is evaluated using standard performance metrics, including accuracy, precision and recall. The proposed system shows promising results and demonstrates potential for integration into clinical workflows for early detection and support in decision-making

Keywords: Endometrial Cancer (EC), Histopathology, Image Preprocessing, CNN, Histopathological Images Machine Learning (ML), Deep Learning (DL).

Abstract: The absence of structured communication between alumni and students continues to limit mentorship, professional networking, and career development opportunities within academic institutions. To address this gap, this paper presents the design and analysis of an Alumni Interaction Platform (AIP)—a centralized web-based system that fosters continuous engagement through mentorship programs, job and internship postings, discussion forums, and real-time communication tools. The proposed platform integrates modern web technologies such as React.js, Node.js, and Firebase to provide a responsive interface, secure data handling, and seamless connectivity. By implementing role-based access control and cloud-based architecture, the system ensures scalability, user privacy, and efficient data synchronization. The study evaluates existing solutions and identifies the technological and social challenges in deploying such systems at institutional scale. This work aims to transform fragmented alumni networks into an organized, technology-driven ecosystem that enhances student outcomes and institutional collaboration.

Keywords: Alumni Interaction, Mentorship Platform, Career Networking, Real-Time Communication, Web Development, React.js, Node.js, Firebase, Role-Based Access Control, Cloud Integration.

Abstract: Predicting the stock market is an extremely difficult endeavor given the unpredictable and nonlinear characteristics of financial markets, with price fluctuations occurring very quickly to complicate forecasting even more. The use of traditional statistical techniques frequently proves inadequate to replicate the complexities of stock movements, which has been the motivation behind increasing attention on the utilization of machine learning methodologies. This research examines the application of Long Short-Term Memory (LSTM) networks to predict stock prices based on past stock prices, using historical stock price data from 2019 to 2023, including Open, High, Low, and Close prices. The performance of the model is assessed through significant metrics like Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE), indicating that LSTM can be used to successfully forecast long-term trends in stock prices. The study also investigates the performance of other machine learning models, including Gated Recurrent Units (GRU) and Artificial Neural Networks, for forecasting stock prices, with the results indicating that LSTM is superior to these models in identifying long-term dependencies. Nevertheless, it remains difficult to predict sudden changes in the market due to externalities, such as economic developments or geopolitical changes. The paper explores possible avenues for future work, such as combining sentiment analysis, hybrid models, and investigating the application of other deep learning architectures to further improve predictive power. The work adds to the body of research on machine learning in financial forecasting and sheds light on how stock market prediction models can be made more robust and accurate.

Keywords: Stock market prediction, machine learning, Long Short-Term Memory (LSTM), Gated Recurrent Units (GRU), Artificial Neural Networks.

Abstract: Legal Sphere is a web-based AI-powered legal assistant designed to simplify access to India’s complex legal system through intelligent automation and multilingual support. Built using Next.js, Type Script, and Tailwind CSS for a responsive interface, and powered by Google Genkit, Firebase, and Large Language Models (LLMs) such as GPT and Gemini, the system delivers real- time, context-aware legal assistance. It allows users to input queries in natural language—English or regional Indian languages—and receive concise, legally relevant responses derived from Indian laws, judgments, and constitutional texts. The architecture integrates secure authentication, fast server-side rendering (SSR), and scalable deployment to ensure performance and reliability. Experimental results demonstrate efficient response times, robust authentication, and improved accessibility to legal information. LegalSphere thus represents a major step towards democratizing legal knowledge, propagating legal literacy, and bridging the justice accessibility gap using artificial intelligence.

Keywords: Legal Technology, Natural Language Processing, AI Legal Assistant, Large Language Models, Next.js, Google Genkit, Multilingual Chatbot, Legal Awareness, Access to Justice.

Abstract: This Project paper present the implementation of digital filters on an FPGA platform for real-time audio signal processing. This utilizes the Basys-3 FPGA board and the peripheral module (PMOD) I2S2 accessory to process audio data using digital finite impulse response (FIR) filters. The extension of the filter kernel length is achieved by converting the dual- channel input to a single-channel output, and the use of free LUTs (Look Up Tables) as digital signal processing (DSP) multipliers, optimizing the utilization of available resources. The successful implementation of the filters highlighted the potential for FPGA- based solutions in audio engineering and digital signal processing. The results provide valuable insights that can guide future work in optimizing FPGA- based digital filters for audio engineering.

Keywords: Digital filter,Field Programmable gate array(FPGA),Digital Signal Processing(DSP).

Abstract: In today’s digital environment, ensuring that academic and professional documents are genuine has become increasingly difficult, especially with the growing incidents of tampering and forgery. The proposed Blockchain-Based Identity Verification System offers a secure and transparent method for issuing and validating digital certificates. By storing certificate information on the blockchain, the system guarantees immutability—once a certificate is recorded with its unique ID, it cannot be altered or replicated [1]. IPFS further enhances the system by providing decentralized and reliable file storage [2]. Authorized institutions can issue certificates digitally, and verifiers such as recruiters or background-check teams can confirm their authenticity through a simple online verification portal, without depending on any intermediaries. This not only strengthens security and trust but also supports a shift toward paperless verification, reducing manual workload and environmental impact. Overall, the system aims to build a dependable digital credential ecosystem that benefits both educational institutions and employers.

Keywords: Blockchain, IPFS (Inter Planetary File System), decentralized storage, Identity verification

Abstract: Seed quality plays a crucial role in ensuring high crop yield and sustainable agriculture. The presence of weed seeds mixed with normal crop seeds reduces germination efficiency, lowers productivity, and increases the cost of weed management. Traditional manual separation of weed seeds is labour-intensive, time-consuming, and prone to errors. To address this challenge, this project proposes a real-time automated weed seed detection system using deep learning. The system employs YOLOv11, a state-of-the-art object detection algorithm, integrated with a Raspberry Pi and camera module for on-field, real-time processing. The YOLOv11 model is trained on a dataset of crop and weed seeds, enabling it to accurately detect and classify weed seeds within seed samples. The Raspberry Pi provides a cost-effective, portable, and low-power platform for implementation, making the system suitable for practical agricultural applications. The proposed solution enhances seed purity assessment by offering high-speed, reliable, and automated detection, ultimately improving crop productivity and reducing dependence on manual labor. This system can be further extended for large-scale seed processing units and integrated with sorting mechanisms for complete automation.

Keywords: Weed Seed Classification, Image Processing, Deep Learning, Raspberry Pi, Camera Module, YOLO v11, Precision Agriculture, Machine Vision.

Abstract: Coding interviews, hackathons held over the inter net, and international software collaborations all necessitate plat forms that provide real-time, interactive coding environments. But most of the available tool lack in providing smooth safe, and multi-language support for effective collaboration. In this paper, we introduce the Real-Time Code Collaborator (RTCC), a browser-based, full-stack system that allows multiple users to code, execute, and debug programs in real time collaboratively. RTCC integrates the strength of Web Sockets for live synchrony, Docker containers for secure running, and a React/Next.js frontend with voice and chat support. The platform not only increases productivity for remote teams but also enables greater accessibility in technical education and hiring. We cover the design principles, technologies, and architecture employed to build RTCC and contrast it with conventional tools such as Google Collab and VS Code Live Share.

Keywords: Real-time coding, Cloud IDE, Web Sockets, Collaborative development, Code execution, Multi Language support, OAuth 2.0, Git integration

Abstract: Water quality decline is a major global issue, leading to economic losses and public health risks. Traditional testing methods are often slow, costly, and geographically limited, highlighting the urgent need for continuous, real-time assessment solutions. This project introduces an innovative Internet of Things (IoT)-based system for real-time water quality detection. It's designed to revolutionize environmental monitoring by integrating advanced sensor technology, embedded systems like the ESP8266. The system uses a comprehensive suite of sensors—including those for pH, turbidity, temperature, dissolved oxygen, and total dissolved solids (TDS)—to provide a holistic, detailed view of water conditions, allowing for the early detection of both gradual changes and sudden contamination events. Data is processed at the edge through embedded firmware, utilizing communication protocols like Wi-Fi to transmit validated data to scalable cloud platforms like Firebase IoT Core. This data then powers a user-friendly interface featuring web and mobile dashboards for real-time data visualization, historical trend analysis, predictive alerts, and remote configuration. Ultimately, this system offers significant improvements over conventional methods in terms of cost, responsiveness, and geographical coverage, providing essential tools for effective water quality management, public health protection, and efficient water resource utilization.

Keywords: ESP8266 microcontroller, Real-time monitoring, Ph, TDS, Turbidity, Alert system, Buck converter, Dashboard, Firebase

Abstract: In an era of increasing digital connectivity, the sophistication and frequency of cyberattacks have grown exponentially, rendering traditional rule-based intrusion detection systems (IDS) insufficient. This literature survey explores the recent advancements in AI-powered IDS solutions, with a particular focus on machine learning (ML)-driven approaches for harmful packet detection. The review analyzes 25 recent research papers published between 2020 and 2025, highlighting trends in model development, dataset utilization, real-time deployment, edge computing, and automation in threat response. While many existing systems achieve high detection accuracy using algorithms such as Random Forest, SVM, CNN, and ensemble techniques, they often fall short in critical areas—such as real-time performance, attack simulation, automated remediation, and handling minority class attacks. This survey identifies those gaps and establishes the motivation for a lightweight, modular IDS that not only detects but also responds to intrusions through intelligent patch recommendations. By comparing existing approaches and their limitations, the paper lays the foundation for building adaptive, scalable, and semi-autonomous security solutions suitable for modern network environments.

Keywords: Intrusion Detection System, Machine Learning, NSL-KDD, Network Security, Automated Patching, Real-Time Threat Detection, Cyberattack Classification, Lightweight IDS

Abstract: The Child Health Tracker is a web-based application designed to assist parents and caregivers in monitoring the health, nutrition, and overall development of children. Managing child health data such as growth, sleep, food intake, vaccinations, and medications is often fragmented across multiple sources, making it difficult to identify patterns or potential health concerns. The proposed system integrates modern web technologies with AI-driven analysis to provide a centralized and intelligent childcare platform. The system is developed using a React frontend and a Node.js Express backend, with Indexed DB used for local data storage to enable offline access. AI services are integrated to analyze meal images, provide nutrition recommendations, and generate health insights. Interactive dashboards and visualizations help caregivers understand growth trends and health patterns easily. The Child Health Tracker aims to simplify child healthcare management, promote healthy habits, and support informed decision-making without requiring complex backend infrastructure.

Keywords: Child Health Tracker, AI-based Recommendation, React, Indexed DB, Health Monitoring

Abstract: Electronic Health Record (EHR) systems face critical challenges in security, interoperability, and patient data sovereignty. Centralized databases remain vulnerable to ransomware attacks and data breaches, while siloed systems prevent seamless provider communication. This paper presents a fully-functional, production-grade prototype integrating blockchain technology, decentralized storage via IPFS, and generative AI for intelligent clinical decision support. The system employs a novel four-layer architecture providing multi-persona web interface supporting patients, doctors, and hospital administrators, Node.js API server for orchestration and authentication, decentralized persistence layer combining Ethereum smart contracts with IPFS for scalable off-chain storage, and Google Gemini AI for real-time clinical analysis. The core innovation presents an AI-assisted prescription workflow analyzing physician drafts against complete patient medical history to generate drug-drug interaction warnings, contraindication alerts, and dosage recommendations. The system demonstrates superior security through blockchain immutability, enhanced interoperability via decentralized architecture, and improved clinical outcomes through context-aware AI analysis. The framework successfully addresses longstanding healthcare IT challenges while maintaining physician autonomy through human-in-the-loop design principles.

Keywords: Blockchain, Healthcare, Electronic Health Records, Artificial Intelligence, Clinical Decision Support, IPFS, Smart Contracts, Decentralized Applications, Medical Data Management, Web3.

Abstract: MedGuard Edge is a decentralized healthcare IoT system that ensures continuous and secure patient monitoring while solving the fundamental security and privacy concerns of traditional centralized systems. The smart wearable hand band includes sensors for temperature, oxygen saturation (SpO2), heart rate (BPM), and humidity, which process crucial biomedical data. These data are encrypted and transferred to the MedGuard server via a Node MCU module for threat analysis, anomaly identification, and decision-making, with the user receiving emergency notifications. At its core, Clustered Federated Learning allows for local model training on clustered devices without exchanging raw patient data, hence ensuring privacy. Blockchain technology secures model updates via tamper-proof validation, ensuring data integrity. Real-time anomaly detection monitors devices and data for anomalies, while self-healing features isolate or recover compromised nodes to ensure system stability. A real-time dashboard displays graphical views of patient data, alerts, device health, and blockchain logs, allowing healthcare administrators to monitor and respond more efficiently.

Keywords: Blockchain, Clustered Federated Learning, Internet of Things (IoT), Healthcare Security, Anomaly Detection, and Self-Healing Systems.

Abstract: The growing need for high-precision, customizable, and rapid manufacturing has led to significant advancements in additive fabrication technologies. This project focuses on the development of a next-generation 3D printer designed to deliver superior accuracy, reliability, and material versatility while maintaining low operating noise. The proposed system features a reinforced mechanical frame and optimized motion control architecture to minimize vibration and increase dimensional stability during printing. A multi-filament feeding mechanism enables seamless compatibility with three widely used materials—PLA, ABS, and Nylon—allowing users to fabricate functional prototypes with varying mechanical properties. The printer integrates Wi-Fi connectivity for wireless control, real-time monitoring, and remote parameter adjustment, enhancing usability and workflow efficiency. Noise-optimized cooling systems, insulated stepper drivers, and vibration-dampening components contribute to significantly reduced acoustic output, making the printer suitable for indoor and educational environments. The firmware is customized to support adaptive slicing, automatic calibration, and intelligent fault detection. Experimental results demonstrate improved print quality, material flexibility, and operational convenience compared to conventional desktop FDM printers. These enhancements validate the system as a robust, modular, and future-ready platform for prototyping, research, and small-scale manufacturing.

Keywords: Printing, Additive Manufacturing, Multi-Filament System, Low-Noise Operation, Wi-Fi Connectivity, FDM Technology, PLA/ABS/Nylon, Smart Prototyping

Abstract: Efficient donor–recipient matching is a critical step in organ transplantation, yet most hospitals still depend on manual comparison of clinical factors such as age, blood group, comorbidities, and organ-specific health indicators. This manual process is slow, prone to inconsistency, and unsuitable for handling the rapid inflow of medical data in real-world environments. To overcome these challenges, this study introduces an intelligent, machine-learning–enabled matching system designed to provide fast, reliable, and data-driven compatibility predictions. The proposed web-based framework incorporates Random Forest and K-Nearest Neighbours models along with computed clinical metrics—including a compatibility score, an organ function score, and a consolidated match score—to evaluate donor–recipient pairs for heart, kidney, liver, and lung transplants. The platform integrates role-based interfaces for administrators, doctors, receptionists, and patients, ensuring streamlined data entry, treatment management, and prediction access. Experimental analysis shows that the system delivers accurate compatibility assessments with efficient real-time execution, demonstrating the potential of machine learning to minimize mismatches, shorten waiting periods, and enhance clinical decision support in transplant workflows. The modular architecture also supports future expansion to additional organs and evolving hospital datasets.

Keywords: Organ Transplantation, Donor–Recipient Matching, Machine Learning, Random Forest Classifier, K-Nearest Neighbours (KNN), Compatibility Prediction, Clinical Decision Support System, Medical Data Processing.

Abstract: Rheumatoid Arthritis (RA) is a chronic autoimmune disease that causes progressive joint inflammation and irreversible structural damage if not diagnosed at an early stage. Accurate classification of RA subtypes, such as seropositive and seronegative RA, along with the identification of erosive joint changes from radiographic images, is essential for effective clinical decision-making. However, conventional diagnostic approaches rely heavily on manual interpretation of laboratory biomarkers and X-ray images, which are time-consuming and subject to inter-observer variability. This work proposes an artificial intelligence–based dual-modal diagnostic framework for automated rheumatoid arthritis disease subtype classification. The system integrates numerical clinical biomarkers and hand X-ray imaging to provide complementary diagnostic insights. The numerical model utilizes six key laboratory parameters—age, gender, rheumatoid factor (RF), anti-cyclic citrullinated peptide (Anti-CCP), C-reactive protein (CRP), and erythrocyte sedimentation rate (ESR)—to classify patients into Healthy, Seropositive RA, and Seronegative RA using an XGBoost classifier. In parallel, an EfficientNet-B3 deep learning model is employed to analyze hand X-ray images for erosive and non-erosive joint damage detection. Experimental evaluation demonstrates that the numerical model achieves an accuracy of 89.28% with a ROC-AUC of 93.21%, while the imaging model attains 85.83% accuracy with a 95.04% recall for erosive cases. The proposed system is deployed as a real-time, web-based clinical decision support tool using Streamlit, providing fast and interpretable predictions. This approach highlights the effectiveness of multimodal AI systems in enhancing early RA diagnosis and subtype classification.

Keywords: Rheumatoid Arthritis, Disease Subtype Classification, Medical Imaging, Machine Learning, Deep Learning, Clinical Decision Support

Abstract: This project presents a machine learning-driven fruit profiling system utilizing advanced deep learning and computer vision techniques to analyse fruit images. The system comprises two main components: fruit type identification with caloric estimation, and fruit ripeness classification. The first component detects various categories of fruits providing estimated nutritional information based on recognized types. The second component assesses the ripeness stage, distinguishing different maturity and spoilage levels across multiple fruit varieties. Both components employ the YOLO V9 algorithm for accurate and efficient detection. By integrating static nutritional data with dynamic quality assessment, the system offers a comprehensive tool for evaluating produce through image analysis. This approach enables quick, automated classification and quality estimation, facilitating applications in nutrition tracking, agricultural management, and supply chain monitoring.

Keywords: Fruit profiling, deep learning, computer vision, YOLO V9, fruit classification, caloric estimation, ripeness detection, image analysis, nutritional assessment, produce quality, machine learning, object detection.

Abstract: Brain–Computer Interfaces (BCIs) enable direct communication between human cognitive activity and electronic systems without requiring physical movement. Recent advancements in wearable EEG technology, microcontroller processing capabilities, and low-cost sensor fusion techniques have opened new pathways for neuro-controlled robotic systems. This research presents the development and validation of a non-invasive, EEG-triggered nano-drone lift-off system using the Neuro-Sky MindWave Mobile 2 headset and an ESP32 microcontroller. EEG Attention signals are acquired over Bluetooth using the Think Gear protocol, decoded on the ESP32, and passed through a carefully tuned threshold-based decision algorithm that activates the propulsion system. A complementary filtering method, combined with a PID-based stability controller, integrates real-time orientation data from an MPU6050 6-axis gyroscope–accelerometer to ensure stable flight. Two DRV8833 dual-channel motor drivers drive four brushed DC motors attached to a custom 3D-printed nano-quadcopter frame. The system successfully demonstrates hands-free lift-off triggered purely through cognitive focus, achieving reliable activation, stable hover behaviour, and low-latency control. The prototype proves that low-cost, single-sensor EEG systems can be effectively integrated with autonomous nano-drones, paving the way for scalable, accessible neuro-controlled robotic platforms. This work has implications in assistive technology, rehabilitation robotics, telepresence applications, and future human–machine interaction research.

Keywords: Brain Computer Interface, Neuro-Sky, EEG, Drone Control, Assistive Technology, Signal Processing, Real-Time Systems.

Abstract: Urban air pollution represents a significant public health challenge where traditional Continuous Ambient Air Quality Monitoring Stations (CAAQMS) provide accurate measurements but suffer from sparse spatial distribution. This research presents an integrated framework combining mobile IoT sensors with hybrid deep learning for comprehensive air quality assessment. The system deploys ESP32-based sensor modules with electrochemical gas detectors (MQ-135, MQ-7, MQ-136) and optical particulate matter sensors to capture spatially distributed measurements of PM2.5, NO2, CO, and SO2. A hybrid CNNLSTM model processes spatial patterns and temporal dependencies to calibrate sensor readings and generate Air Quality Index (AQI) forecasts. The prototype implementation demonstrates feasibility, achieving Mean Absolute Error of approximately 24 AQI units, with complete mobile deployment projected to reduce errors by 20-40% and provide city-wide coverage with over 50,000 daily measurements.

Keywords: Air Quality, Deep Learning, IoT Sensors, Forecasting, Sensor Calibration

Abstract: The efficient management of student attendance is a persistent administrative challenge in educational institutions. Conventional manual and token-based attendance systems are time-consuming, error-prone, and vulnerable to proxy attendance. Although biometric solutions such as fingerprint scanners address identity verification, they introduce hygiene concerns and operational bottlenecks. This paper presents a real-time, contactless, and fully automated student attendance monitoring system based on computer vision and deep learning techniques. The proposed system integrates YOLOv8 for high-speed face detection with the VGG-Face model for robust facial recognition. A novel duration-based attendance validation mechanism is introduced, wherein a student is marked present only after being continuously or cumulatively recognized for a predefined duration. The system further automates attendance reporting through Excel generation and real-time email notifications using SMTP. Experimental evaluation demonstrates high accuracy, robustness to occlusion and lighting variations, and suitability for real-world classroom deployment.

Keywords: Automated Attendance, Face Recognition, YOLOv8, VGG-Face, Deep Learning, Computer Vision

Abstract: The rapid expansion of the Internet of Things (IoT) presents a viable solution to inefficiencies in traditional brick-and-mortar retail, particularly the problem of lengthy checkout queues [7, 8]. This paper details the design and implementation of a cost-effective, IoT-enabled smart shopping cart that addresses key practical limitations found in existing systems. Our proposed design incorporates a novel **Hybrid Scanning Mechanism**—integrating both a Radio Frequency Identification (RFID) reader and a Non-RFID (Barcode/QR) reader to ensure universal compatibility with both tagged and untagged supermarket items. Critically, the system introduces a **Real-Time Weight Sensor (Load Cell)** to perform robust anti-theft validation by matching the physical weight of items with the accumulated digital bill, preventing item substitution. The system utilizes an Arduino microcontroller, an ESP module for wireless connectivity, and Bluetooth for communication with a customer’s mobile device, achieving the primary objectives of automated billing, queue reduction, and enhanced transaction security. Experimental validation demonstrates that the hybrid approach offers a practical, scalable, and secure alternative to current single-technology smart cart prototypes.

Keywords: IoT, Smart Shopping Cart, RFID, Load Cell, HX711[ESP module], MFRC522[Wi-Fi module].

Abstract: Lung cancer is a major health concern and should be predicted at an earlier and precise stage to enhance outcomes for patients. This project offers a hybrid machine learning approach by integrating the analysis of symptom-based surveys and the analysis of CT scan images for risk prediction of lung cancer. Patient complaints are analyzed by a rule-based weighted scoring system to obtain a preliminary result of risk levels associated with the possibility of lungs being affected by cancer. The CT scan images are pre-processed by techniques involving the resizing and sharpening of images, applying threshold levels, and edges for extracting significant details, which are then processed by ResNet50 for extracting the extracted featured details by a Convolutional Neural Network (CNN).

Keywords: Machine Learning, CNN, ResNet50, Image Preprocessing.

Abstract: In today's fast-paced world, interacting with computers using traditional input devices like keyboards and mice can be cumbersome and inefficient. This project presents the design and implementation of a voice- activated assistant built using the Python programming language. Leveraging libraries such as speech recognition and text-to-speech, the assistant listens to user voice commands, converts speech into text, interprets the intent, and executes predefined tasks — such as opening applications or websites, fetching the current time or date, performing web searches, delivering news or weather updates, and managing simple tasks like reminders or basic arithmetic. The system aims to provide a hands-free, accessible, and user-friendly interface, especially benefiting users who have difficulty with conventional input or who prefer voice interaction. By reducing dependency on hardware input devices and automating routine actions, the assistant enhances user convenience and productivity. The modular architecture ensures extensibility and allows future integration of more advanced natural language processing or machine learning components for smarter responses.

Keywords: Voice Assistant, Speech Recognition, Text-to-Speech, Natural Language Processing (NLP), Virtual Assistant

Abstract: Cryptography has evolved from primitive methods of secrecy into a mathematically rigorous discipline that underpins privacy, trust, and security in the modern digital ecosystem. This paper presents a comprehensive exploration of cryptography as both a technical and socio-political construct, tracing its historical progression from classical ciphers and mechanical encryption devices to contemporary digital and post-quantum cryptographic systems. The study examines the foundational mechanisms of modern cryptography, including symmetric encryption, asymmetric key cryptography, hashing, and digital signatures, highlighting their complementary roles in ensuring confidentiality, authentication, and data integrity. Special emphasis is placed on hybrid cryptographic architectures such as Transport Layer Security (TLS), which form the backbone of secure internet communication. Beyond technical foundations, the paper analyzes cryptography’s role as a protector of human rights, particularly privacy and freedom of expression, and discusses the enduring “Crypto Wars” surrounding lawful access and encryption backdoors. Emerging privacy-enhancing cryptographic techniques, including Zero-Knowledge Proofs and Homomorphic Encryption, are evaluated for their potential to enable secure computation without data exposure. Finally, the paper addresses the existential threat posed by quantum computing to current cryptographic standards and outlines the urgent transition toward post-quantum cryptography, with a focus on NIST-selected lattice-based algorithms. The paper concludes that cryptography is not merely a security tool but a fundamental pillar of digital trust, requiring continuous innovation, sound policy, and global cooperation to safeguard future information systems.

Keywords: Cryptography, Digital Security, Symmetric Encryption, Asymmetric Cryptography, AES, RSA, Elliptic Curve Cryptography, TLS/SSL, Digital Signatures, Privacy-Enhancing Cryptography, Zero-Knowledge Proofs, Homomorphic Encryption, Crypto Wars, Human Rights, Post-Quantum Cryptography, Quantum Computing

Abstract: This research paper presents a comprehensive app-based Attendance Management System developed using Kotlin, Firebase, and the Model View View Model (MVVM) architecture. The system simplifies the process of recording and analyzing student attendance in educational institutions by leveraging modern mobile technologies and cloud services. By utilizing Jetpack Compose for the frontend user interface and Firebase for backend data storage and authentication, the system provides a reliable, responsive, and real-time platform for teachers, students, and administrators. The application significantly reduces manual workload, eliminates common errors associated with traditional paper-based methods, and enhances institutional transparency through advanced data visualization, reporting features, and automated notifications. It incorporates secure authentication mechanisms, real-time synchronization, automated email and push notifications for low attendance alerts, and exportable reports in various formats, addressing modern educational challenges effectively. Our analysis combines software engineering principles with empirical evaluation to demonstrate how digital transformation can optimize administrative processes in education. Results from deployment in a test environment indicate over 99% accuracy in attendance recording, a 70% reduction in administrative time, and improved student engagement with attendance tracking. The system shows promise for scalability across different educational levels, with potential integration into larger learning management systems.

Keywords: attendance management system, mobile application, Kotlin, Firebase, MVVM architecture, Jetpack Compose, educational technology, real-time database, data visualization, automated notifications

Abstract: Ash is an intelligent, voice-enabled desktop assistant designed to enhance user productivity and automate complex tasks through natural language processing and machine learning. The system captures user voice commands and text inputs, processes them using advanced NLP techniques, and executes a wide range of desktop automation tasks including opening/closing applications, system controls (lock, sleep, alarm, volume adjustment), web searches, and real-time information retrieval. Additionally, Ash features a local image generation module that creates images based on user prompts, enabling creative content generation directly from voice or text commands without requiring external API calls. The system integrates reinforcement learning for adaptive user behavior prediction, context-aware voice tone adjustment. Unlike commercial assistants, Ash provides transparent, fully-customizable implementation with complete developer control.

Keywords: Virtual Assistant; Desktop Automation; Speech Recognition; Natural Language Processing; Intelligent User Interface; Task-Oriented Dialogue System.

Abstract: Ensuring the safety of women has become an urgent issue in contemporary society, as instances of harassment, abuse, and violence continue to appear in both physical and virtual spaces. Despite increasing awareness and evolving legal frameworks, many cases remain unreported due to fear, embarrassment, or a lack of accessible and reliable support systems. Mahila Suraksha Nyayavani is an intelligent web-based application developed to provide a secure and confidential environment for women to report incidents safely. The system incorporates a Natural Language Processing (NLP) model built on BERT (Bidirectional Encoder Representations from Transformers) to automatically interpret and classify reports based on their level of urgency and potential risk. Critical cases are prioritized for faster response and intervention by concerned authorities. In addition to its reporting function, the platform emphasizes awareness and empowerment through categorized articles, self-defense tutorials, and real-life case studies. It also includes a chatbot interface that provides instant assistance, step-by-step guidance, and legal information. By integrating AI-based risk evaluation, digital awareness, and institutional collaboration, Mahila Suraksha Nyayavani acts as a holistic safety and education platform, strengthening both prevention and response mechanisms for women’s security.

Keywords: Women Safety, Crime Reporting, NLP, Sentiment Analysis, Deep Learning, BERT, Risk Prediction, Web Application.

Abstract: Diabetic Retinopathy (DR) is a severe microvascular complication of diabetes and one of the leading causes of preventable blindness worldwide. Early diagnosis and accurate classification of DR stages are essential for timely medical intervention and effective treatment. However, traditional manual screening of retinal fundus images by ophthalmologists is time-consuming, subjective, and prone to inter-observer variability. This paper presents an automated system for the Analysis And Classification Of Diabetic Retinopathy using Deep Learning techniques. The proposed approach utilizes retinal fundus images collected from standard public datasets such as Kaggle and clinically sourced datasets. Image preprocessing and enhancement techniques are applied to improve retinal feature visibility, followed by deep feature extraction using a Convolutional Neural Network (CNN) based on the ResNet50 architecture. The trained model classifies retinal images into distinct DR severity stages, including No Diabetic Retinopathy, Mild Diabetic Retinopathy, and Severe Diabetic Retinopathy. Experimental evaluation demonstrates that the proposed system achieves high classification accuracy, sensitivity, and specificity, making it suitable for real-world screening applications. A user-friendly graphical interface is also developed to assist clinicians by providing rapid and reliable DR predictions. The proposed system serves as an effective computer-aided diagnosis tool, reducing screening workload and improving early detection of diabetic eye diseases.

Keywords: Diabetic Retinopathy, Deep Learning, Convolutional Neural Networks, ResNet50, Medical Image Analysis, Retinal Fundus Images, Computer-Aided Diagnosis

Abstract: Floods, earthquakes, cyclones and landslides commonly destroy essential communication facilities like mobile towers, fibre links and internet services. This breakdown prevents those who are in need from sending out an SOS or sharing their current location which greatly slows down the rescuers and causes more people to be killed. In order to solve these problems, I will use AI technology to design an emergency rescue communication system that can provide continuous, stable and reliable communication for both online and offline environments. When the network is available, it uses GPS and cloud to send exact location data along with structured emergency request types like Ambulance, Water Supply, Resources, General Help etc. Without Internet Services: In the absence of internet service, the system smartly forms a wireless mesh network among adjacent cell phones that enables multi-hop messages to travel over devices connected via a link in between. An integrated AI-based routing module considers battery level, RAM availability, storage space, and signal strength; selects the best & most energy-efficient communication path for sending a message, using ML algorithm to increase the reliability of delivering a message under harsh environment. A real-time rescue dashboard, with color codes of (Red, Yellow, Green) status, provides the rescue team members with better visibility and understanding for a faster interpretation and better coordination in times of emergency. The Proposed System increases Situational Awareness & Reduces Gaps on Communication, so it can give Timely help as a Scalable & Flexible Life Saver framework for Modern Disaster Management Operations

Keywords: Emergency Communication, Wireless Mesh Network, Disaster Management, Reinforcement Learning, AI Routing, Offline Communication, Rescue System.

Abstract: Smart Reader is designed as an accessibility-focused system that supports non-visual reading by combining text recognition, scene understanding, audio guidance, and tactile cues. The platform utilizes a pair of complementary OCR engines in concert with a lightweight object-analysis module to interpret both text and surrounding context from incoming video frames. Such an optimized backend, implemented on top of Fast API, empowers the pipeline to handle each frame within about two hundred milliseconds, which can enable several frames per second on regular CPU hardware. A specially designed mapping function transforms the OCR characteristics, including confidence levels, text density, and semantic weight, into structured vibration signals that help users haptically perceive document layout. Experimental studies have shown higher recognition reliability than using a single OCR method alone, along with consistent detection quality and responsive system behavior. In general, Smart Reader provides an improved pathway for visually impaired users to access printed or on-screen information based on a powerful combination of perception, interpretation, and haptic assistance.

Keywords: OCR, Haptic Feedback, YOLOv8, Assistive Technology, Fast API, Real-Time Processing.

Abstract: EEG-based object recognition is gaining attention as brain signals provide unique neural patterns when visual stimuli are perceived. This research proposes an automated classification pipeline that learns EEG temporal dependencies using a 1D Convolutional Neural Network (1D-CNN). EEG signal segments collected from five electrode positions—AF3, AF4, T7, T8, and PZ—are integrated to construct a spatial feature matrix containing diverse signal responses. Feature normalization is applied using standard statistical scaling to maintain consistent input distribution, and object Object categories are converted into numeric class identifiers for multi-class model training. class indices to support multi-class learning. The The model structure is composed of multiple processing layers designed for EEG pattern learning. contains sequential 1D convolutional layers that capture short-range temporal interactions, followed by max-pooling to reduce noise sensitivity and support stable feature extraction. Dense layers further learn high-level signal abstractions, leading to a A softmax output layer that converts raw scores into a normalized probability distribution probability-based classification. Training is performed using an 80:20 data split using batch-driven learning to stabilize gradient updates. For end-user inference, the trained model is deployed on the Hugging Face cloud using a Gradio interface to support real-time prediction and confidence visualization through a dynamic gauge chart.

Keywords: Brain-Computer Interface, EEG Signal Classification, 1D Convolutional Neural Network, Visual Stimuli Recognition

Abstract: Exploring the domain of Artificial Intelligence and its utilization towards designing better solution for our tasks highlight the key technological advancement. With these systems using the large language models, machine learning algorithms and natural language processing along with real time data integration an leveraging these for our AI-powered trip planner to address the issue of improper planning and management of the tour, this travel planner considers several factors and constraints such as budget, location, real time considerations to plan out user centric itineraries. Personalization is the key element which helps to optimize the plan and adapt the user behavior for interest and tailor the plan according to the preferences. Multi agent architecture helps to provide a diverse planning considering real time forecasting updates and keeping the localized touch to the travel plan to maintain the dynamic adaptability and feasibility of the plans. With certain data and recurrent planning these AI driven planners evolve as a successful handy solutions than traditional time-consuming plans. Although these AI driven plans also come with certain challenges to overcome such as data privacy, accuracy of the information, real time data driven forecast and user emergencies, integrating highly developed and emerging technologies such as augmented reality, blockchain can be wonderful to enrich the user experience and for long term adaptive learning would be helpful to build a trustworthy AI driven travel solutions.

Keywords: Personalized Travel Planning, Dynamic Itinerary Adaptation, Behavioral Prediction and User Modeling, User Engagement and Feedback Loops, Scalability and Robustness in Travel Planning, Graph-structured Context Management.

Abstract: Counterfeiting and supply chain inefficiencies, especially in the agricultural sector, lead to major safety risks and eco nomic losses. Traditional tracking systems using barcodes, RFID, or centralized databases remain vulnerable to manipulation and lack full transparency. This paper presents AGRO-TRUST!!, a platform integrating Blockchain, IoT, cloud computing, and cyber security to ensure secure and real-time traceability. Each product receives a blockchain-linked QR code, while IoT sensors such as GPS, MQ gas sensors, and DHT11 modules continuously record environmental and location data throughout the logistics cycle. Smart contracts automate secure payment procedures and product verification, while cloud computing enables scalable data processing and connection. Strong cyber security measures prevent manipulation and unauthorized access, such as encryption and role-based access control. 99.5% integrity in IoT data transmission and immediate identification of counterfeit efforts employing cloned QR codes are demonstrated by experimental results. AGRO-TRUST!! offers a transparent, tamper-proof, and scalable solution for agricultural and related supply chains.

Keywords: Blockchain, Internet of Things (IoT), Supply chain, smart contracts, blockchain, IoT, QR codes, and counterfeit detection

Abstract: Face-swap deepfakes present significant challenges to digital media authenticity and have emerged as critical threats to information integrity in contemporary society [1]. This paper proposes an AI/ML-based detection framework that combines Vision Transformer (ViT) feature extraction with facial symmetry analysis through an early fusion architecture [2]. Our approach leverages the Data-Efficient Image Transformer (DeiT-small) backbone [3] to extract high-level visual features, which are concatenated with 50-dimensional facial symmetry metrics computed from 68-point facial landmarks detected using dlib. The fused features (434 dimensions) are classified through a lightweight fully connected layer optimized with cross-entropy loss. Extensive evaluation on a dataset of 140,002 [4] training samples demonstrates robust detection performance with confidence scores exceeding 94% on test samples. The proposed architecture significantly reduces computational overhead compared to multi-stream approaches while maintaining discriminative power through complementary feature modalities. Furthermore, we present a user-friendly Gradio-based web interface [5] enabling practical deployment and batch analysis capabilities. Our results indicate that the synergistic combination of transformer-based visual perception and geometric facial constraints provides an effective solution for face-swap detection in real-world deployment scenarios.

Keywords: Deepfake detection, Face-swap, Vision Transformers, Feature fusion, Facial symmetry, Digital forensics, Web deployment.

Abstract: Water scarcity continues to intensify due to growing population pressure and inefficient household consumption practices, highlighting the need for accessible digital solutions that promote responsible water use. This work presents a full-stack, web-based digital knowledge-sharing platform developed to educate users on practical water-efficient techniques and encourage sustainable consumption behavior. Designed in a wiki-style format, the system organizes conservation methods into structured categories, enabling users to easily discover and apply relevant solutions in daily life. To enhance interactivity, the platform integrates a rule-based water usage calculator that estimates household consumption across common activities and identifies high-usage areas, offering targeted conservation suggestions. The system is implemented using Python Flask for backend logic, SQLite for structured content storage, and responsive web technologies for the user interface. Testing results indicate reliable functionality and effective user engagement, demonstrating that software-driven knowledge platforms can significantly contribute to water conservation awareness and informed decision-making.

Keywords: Water Conservation, Digital Knowledge Platform, Web Application, Sustainability, Water Usage Calculator, Environmental Awareness

Abstract: Millions of people around the globe are impacted by neurological disorders including Alzheimer’s and Parkinson’s, epilepsy and multiple sclerosis, and they have a significant impact on the patients’ quality of life. Early and proper diagnosis is very important in enhancing the treatment outcomes. This study presents a deep learning-based system to automatically detect and identify neurological diseases of the brain imaging images and clinical data. The system combines convolutional neural networks (CNN) with feature extraction and recurrent neural networks (RNN) with time data analysis, and its accuracy was 94.2% when it was used in classification tasks on several datasets. A comparative study of the traditional machine learning models shows better sensitivity and robustness. The research helps in improving the efficiency of diagnosis in healthcare systems, particularly in resource constrained systems.

Keywords: Deep Learning, Neurological Disorders, Brain Imaging, CNN, RNN, Feature Extraction, Medical Diagnosis.

Abstract: Railway track failures such as cracks, misalignment, and obstacles pose serious safety risks and often remain undetected due to the limitations of manual inspection methods. To address this issue, this paper presents an Internet of Things (IoT) based railway track fault detection system that enables continuous and real-time monitoring using intelligent video analysis. A Wi-Fi enabled camera mounted on a moving inspection unit captures live video of the railway track, which is processed using computer vision techniques and a YOLO-based deep learning model to identify structural defects. The detection results are transmitted to a cloud platform using Firebase, allowing remote monitoring and instant alert generation. An ESP32 microcontroller retrieves cloud commands to automatically control the movement of the inspection unit through a relay-driven motor mechanism, ensuring immediate stoppage upon fault detection. The proposed system minimizes human intervention, improves detection accuracy, and offers a cost-effective and scalable solution for enhancing railway safety and enabling predictive maintenance.

Keywords: Railway Track Fault Detection, Internet of Things (IoT), Computer Vision, YOLO, ESP32, Real-Time Monitoring, Cloud Computing.

Abstract: In today’s fast-paced digital world, individuals face increasing challenges in managing their personal wardrobes efficiently. Despite owning a variety of clothing, many struggle with repetitive outfit choices, underutilization of their wardrobe, and daily decision fatigue. DrobeDex is an AI-powered smart wardrobe and outfit planner Progressive Web Application (PWA) designed to address these challenges by combining user-centric design with intelligent automation. The application allows users to digitize their clothing collections through image uploads, which are then auto-tagged using a custom ResNet50 model. Users can create, manage, and log daily outfits using a drag-and drop interface. DrobeDex is built using React.js to ensure cross platform compatibility and smooth performance on modern web browsers. It employs efficient client-side image preprocessing and integrates a custom ResNet50 model for auto-tagging alongside Generative AI for personalized outfit recommendations. The project follows an Agile development methodology, enabling iterative improvement based on user feedback and real-world testing. Index Terms—AI-powered wardrobe, outfit planning, deep learning, React Native, sustainability, personalized recommendations.

Keywords: AI-powered wardrobe, outfit planning, deep learning, sustainability, personalized recommendations.

Abstract: The Accident Detection and Alert System is designed to provide real-time accident detection using deep learning techniques, specifically the YOLOv8 model. This system aims to enhance the speed and accuracy of accident detection through image processing. The platform is developed with a simple, user-friendly interface using HTML, CSS, and JavaScript for frontend development, enabling users to easily log in, register, and interact with the prediction page. Once a user successfully logs in, they are directed to the prediction page, where they can upload an image. The uploaded image is then processed by the backend, which is powered by Python and the Flask framework. The YOLOv8 model is responsible for analysing the image and detecting if an accident is present. YOLOv8, being a fast and accurate deep learning-based object detection model, is ideal for real-time applications like accident detection, as it can process images quickly while maintaining high accuracy. If the model detects an accident in the uploaded image, the system automatically triggers an email alert to the admin. The email contains essential details such as the image and a notification of the accident, allowing the admin to take immediate action. If no accident is detected, the user is notified via the interface, informing them that no accident was found in the image.The system's workflow aims to provide efficient and reliable detection of accidents, reducing response times in emergency situations. It can be applied in various real-time monitoring scenarios, including traffic management, surveillance systems, and emergency response systems. By automating the accident detection and alerting process, the system enhances communication between users and admins, ensuring that the right actions are taken as quickly as possible. The YOLOv8 model ensures that the detection process remains both fast and accurate, making the solution effective for use in dynamic, high-demand environments where real-time responses are critical.

Keywords: Accident Detection, YOLOv8, Flask, Real-Time Image Processing, Email Notification, Python, Machine Learning, Traffic Monitoring, Admin Alert, Image Upload, Web Interface, Emergency Response, Security Systems, Object Detection.

Abstract: The interpretation of vast genomic datasets remains challenging due to complexity and cognitive burden on clinicians. The NextGen AI Genomic Biomarker System addresses these challenges through a hybrid architecture combining NLP and Deep Learning. The system leverages TF-IDF vectorization with Random Forest classification achieving weighted F1-score of 0.874, and employs CNN-LSTM architecture achieving AUC of 0.93. Integrated with SHAP-based explainability, the system provides transparent predictions with sub-2-second latency while maintaining HIPAA/GDPR compliance. Index Terms—Genomic biomarkers, precision medicine, NLP, deep learning, explainable AI, Random Forest, CNN-LSTM.

Abstract: Online users are increasingly exposed to malicious websites disguised as legitimate ones, aiming to steal sensitive information. To combat these threats, PhishGuard, a network intrusion detection system, analyzes URLs using machine learning to classify sites as safe or malicious. It examines features such as domain structure, URL length, special characters, and domain age to detect phishing attempts accurately and in real-time.It is designed to be efficient with low false positives and scalable for future enhancements, providing robust protection against modern cyber threats Index Terms: Phishing Detection,URL Feature Extraction,Real-Time Detection,Cybersecurity,Malicious URL Classification, Scalability, False-Positive Reduction.

Abstract: Open-source software is extensively used in modern systems due to its flexibility and cost efficiency; however, it often contains hidden security vulnerabilities that traditional testing methods may fail to detect. Fuzz testing is an automated technique that addresses this challenge by supplying programs with random and malformed inputs to uncover crashes and weaknesses. This paper presents a web-based system that demonstrates how fuzzing improves the security of open-source files. The system allows users to upload single files, multiple files, or compressed archives and simulates the processes of building, instrumentation, and fuzzing. Security improvements are analyzed using metrics such as code coverage, crash detection, vulnerability count, and overall security score. A comparative evaluation is performed to highlight the difference in software robustness before and after fuzzing. The proposed system integrates an interactive frontend with a FastAPI-based backend to provide real-time progress visualization and automated result reporting. The results indicate that fuzzing significantly enhances the stability and security of open-source files, emphasizing its effectiveness as a proactive software security testing approach.

Keywords: Fuzz Testing, Open Source Software Security, Automated Vulnerability Detection, Software Testing, File Security.

Abstract: Email has become a vital tool due to the growing reliance on digital communication, but because traditional email systems rely on graphical user interfaces and keyboard inputs, visually impaired people experience considerable challenges while utilizing them. The Voice-Based Email System shown in this research allows visually impaired users to send, receive, and navigate emails using only voice instructions. The system creates a n a c c e s s i b l e a n d u s e r - f r i e n d l y communication environment through the employment of Interactive Voice Response (IVR), Text-to-Speech (TTS), and Speech- to-Text (STT) technologies. The system, which is implemented in Python and m a k e s u s e o f m o d u l e s l i k e SpeechRecognition, Pyttsx3, SMTPLIB, and IMAPLIB, guarantees independence, lessens cognitive strain, and improves digital inclusivity for users who are visually impaired.

Keywords: Voice-based email, Speech-to- Text, Text-to-Speech, IVR, Accessibility,Visually impaired users, Assistive technology.

Abstract: Sustainable agriculture requires efficient fertilizer management to enhance crop yield while preserving soil health and the environment. This project presents a Sustainable Fertilizer Usage Optimizer that leverages machine learning and precision farming techniques to recommend optimal fertilizer usage based on soil nutrients, crop type, and environmental conditions. The system integrates crop recommendation, yield prediction, and fertilizer optimization into a unified web-based platform developed using Flask. By minimizing excessive fertilizer application and promoting balanced nutrient management, the proposed solution improves productivity, reduces costs, and supports eco-friendly farming practices. The system empowers farmers with data-driven insights for sustainable and profitable agriculture.

Keywords: Sustainable Agriculture, Fertilizer Optimization, Crop Yield Prediction, Precision Farming, Machine Learning, Soil Nutrient Management, Smart Agriculture, Web-Based Decision Support System.

Abstract: AI-Based System for Skill Mapping and Future Planning is an intelligent digital platform, designed to be simple and personalized for career decision-making by students and young professionals. The system analyzes user resumes using advanced Artificial Intelligence techniques such as Natural Language Processing, BERT-based resume parsing, and Machine Learning models to extract relevant skills, identify existing gaps, and predict suitable career paths aligned with current industry demands. This provides explicit, actionable insights into competencies required for future growth and recommends targeted learning resources through integrated platforms such as YouTube and Udemy. Additionally, the system is embedded with an interactive AI chatbot that will update in real time with career counseling, step-by-step guidance, and motivational assistance, enhancing user engagement and building confidence. The platform is powered by Python, Next.js, and state-of-the-art ML/NLP frameworks that ensure a seamless user experience with fast processing. It minimizes career uncertainty through automation in resume evaluation, skill-gap detection, course recommendation, and job-role mapping, hence democratizing access to personalized career guidance. ProPath represents a significant stride toward narrowing the gap between education and employability, enabling continuous learning and inclusive, sustainable growth in the professional world.

Keywords: Career tech, AI, NLP, skill mapping, machine learning, recommender systems, chatbots.

Abstract: This paper aims to help doctors predict how serious a COVID-19 patient’s condition might become using chest X-ray images and Artificial Intelligence (AI). By analyzing these images with advance deep learning and machine learning techniques, the system can identify patients at high risk early on, allowing doctors to act quickly and prioritize treatment. Key features are selected using smart methods like Principal Component Analysis (PCA), and models such as Bagging, AdaBoost, KNN, and LP Boost have shown excellent performance with up to 97% accuracy. This approach helps hospitals manage resources better and provide timely care to the patients who need it most. Proposed method outperforms the state of art techniques of Covid-19 severity prediction.

Keywords: machine learning, Covid-19 severity deep learning, PCA (Principal Component Analysis).

Abstract: We develop a framework that incorporates clinical information, smoking history and computed tomography (CT) derived radiomics into an artificial neural network (ANN) that can predict early lung cancer risk. We create a multimodal dataset by combining institutional medical record data from LIDC-IDRI images, we extract radiomic features from the images including nodule size, texture entropy, nodule edge sharpness, etc., and we normalize our data through proper imputation and outlier removal techniques and reduce dimensionality of all our extracted data through Principal Component Analysis (PCA). We use a patient split on training data to prevent overfitting in our model and measure performance with several metrics (AUC, Sensitivity, Specificity, Error Inspection through ROC Curve and Confusion Matrix). We pair our predictions with SHAP/LIME based explanations at a case level so that the physician or clinician can identify what variables contributed to their patients' risk scores and assist in developing appropriate thresholds for clinical evaluation. Overall, the combination of our prediction and explanation results provide evidence of the benefits of multimodal ANN risk assessments as well as demonstrate the importance of a transparent and appropriately governed deployment strategy.

Keywords: Lung cancer; risk prediction; artificial neural networks; radiomics; CT imaging; biomarkers; smoking exposure; SHAP; LIME; ROC analysis; confusion matrix; clinical decision support.

Abstract: Food waste keeps piling up, even while so many people struggle to get enough to eat. One major issue? There’s no simple way for people with surplus food to connect with organizations that can deliver it to those in need. That’s where the Food Donation Management System steps in. This digital platform links food donor directly with non-profits and other organizations, streamlining the entire process and cutting out the usual confusion. The system runs on Node.js and Express.js for the backend, with Firebase Firestore as its database. That gives it speed, supports lots of users at once, and keeps everyone’s info updated in real time. Security isn’t just a buzzword here—user passwords are encrypted, so privacy stays protected. There are two main roles: Donors and Organizations. Both can sign up, log in, and interact based on what’s actually needed at the moment. Donors just fill out a few quick details about their cooked meals or groceries, and the platform sends that info right to the right organization. Organizations can browse new donations, accept them, schedule pickups, and track everything. Every step is recorded in a Donation History, so there’s always a clear record. Plus, the system uses SendGrid to shoot out email alerts instantly when a donation is accepted or picked up. All of this cuts down on endless emails and phone calls, closes frustrating communication gaps, and, most importantly, helps make sure good food reaches people—not landfills. It’s easy to use, widely accessible, and truly makes giving back simpler. Really, this isn’t just another tech project. It’s a push for social responsibility and a real move toward tackling hunger.

Keywords: food donation, surplus food management, real-time database, donor–organization coordination, Node.js, Express.js, Firebase Firestore, secure authentication, automated email notifications, SendGrid, food waste reduction.

Abstract: Deepfake detection using machine learning is essential in safeguarding digital content authenticity. This project utilizes CNNs for image analysis, SVMs for audio verification, and Bayesian models for video scrutiny. By refining detection techniques, it ensures reliable identification of manipulated media and enhances digital security against evolving threats. system begins with data preprocessing, removing noise and extracting features essential for analysis. Machine learning models are trained on diverse datasets containing both genuine and synthetic content. Advanced classification algorithms then determine manipulation likelihood, continuously adapting to increasingly sophisticated deepfake generation methods for improved accuracy, By integrating multiple AI techniques, this project provides an automated solution for identifying manipulated content across various multimedia formats. Strengthening digital trust, it addresses growing concerns over misinformation while contributing to ethical AI applications that preserve content integrity, privacy, and authenticity in modern digital communication. The rapid growth of artificial intelligence has made it easier to create convincing fake media, posing serious risks in areas like politics, entertainment, and social media. As fake content becomes more widespread, effective detection methods are crucial. Current approaches struggle to keep up with evolving deepfake technologies, creating a need for reliable solutions. This paper proposes using convolutional neural networks to analyse facial features and motion inconsistencies in videos, aiming to improve detection accuracy. Additionally, audio analysis will be integrated to detect mismatches between sound and visuals, enhancing the model’s effectiveness. The research emphasizes the importance of simple and effective methods to address the challenges of fake media.

Keywords: Deepfake detection, Machine learning, Digital content authenticity, Convolutional Neural Networks (CNNs), Support Vector Machines (SVMs), Image analysis, Video scrutiny, Data preprocessing, Feature extraction, Manipulation detection, Synthetic media, Classification algorithms, Adversarial threats, Multimedia forensics, Digital trust, Misinformation, Ethical AI, Content integrity, Privacy, Facial feature analysis, Motion inconsistencies, Audio‑visual mismatch, Deepfake generation methods, Automated detection system, Reliable detection solutions.

Abstract: The Project named “SKINSCAN - Disease Detection”, a skin-detecting method used to detect skin disease type and its accuracy. This project focuses on using advanced deep learning techniques to accurately classify different skin diseases. It uses a specific model called VGG16, which is great at analysing images. The goal is to develop a reliable model that can automatically identify skin diseases from images, making diagnosis faster and more accurate. -More The dataset used in this project includes five types of skin conditions: Acne-cystic acne, biting fleas, diabetic blisters, spider bites, and vitiligo. The model is designed to recognize these different conditions, ensuring it can handle a variety of skin problems. By using a technique called transfer learning, the pre-trained VGG16 model is finetuned to work with the skin disease dataset. The model goes through extensive training, validation, and testing to ensure it is highly accurate. One of the key successes of this project is the model's high accuracy rate of 98.08%. This means it can correctly identify skin diseases in most cases, which is important for reducing incorrect diagnoses and improving patient care. Additionally, the system can classify skin diseases in real- time, making it a useful tool for doctors and dermatologists. The user interface, created in MATLAB, is designed to be easy to use, allowing healthcare professionals to quickly and accurately make decisions. Overall, this project provides a comprehensive solution for skin disease classification using deep learning, achieving high accuracy and aiding in early diagnosis and effective treatment.

Keywords: Automated Skin Disease Diagnosis, Deep Convolutional Neural Networks, Transfer Learning–Based Classification, dermoscopic Image Analysis, Data Augmentation and Preprocessing, Multiclass Skin Lesion Recognition, Performance Metrics Evaluation, Clinical Decision Support Systems.

Abstract: Because fixed-time and count-based signal control systems have limits, urban traffic congestion continues to be a major problem. VisionFlow, an AI-based intelligent traffic management system that combines adaptive decision-making and real-time computer vision for effective signal optimization, is presented in this study. Using live camera feeds and upstream photos, the system uses the YOLOv8 deep learning model to identify and categorize automobiles. In order to dynamically distribute green signal durations, VisionFlow presents a dual-algorithm architecture that combines an Adaptive Waiting Time (AWT) algorithm with Vehicle Actuated Control (VAC). In contrast to conventional methods, the AWT algorithm ensures equity and less traffic by prioritizing lanes based on both vehicle count and cumulative waiting time. Additionally, the system includes anti-starvation measures, upstream surge detection, and emergency vehicle management. A. Traffic conditions, signal phases, urgency heatmaps, and performance data are all displayed on a real-time interactive dashboard. When compared to VAC, experimental findings show that the suggested AWT technique greatly lowers average waiting time and increases traffic flow efficiency. For next-generation smart city traffic control systems, VisionFlow provides a workable and scalable option.

Keywords: YOLOv8, computer vision, intelligent traffic management, adaptive waiting time algorithm, Actuated Vehicle Control Systems for Smart Cities Optimizing Traffic Signals AI-Powered Traffic Control and Real-Time Traffic Monitoring.

Abstract: As our society becomes increasingly autonomous and utilizes agentic systems, it is important to understand whether these systems subconsciously discriminate against certain populations based on characteristics such as employment status or education level. This research presents a machine learning framework to analyze loan approval decisions while ensuring algorithmic fairness across different demographics. Our study employs a systematic approach by combining multiple classification models with fairness analysis. To address class imbalance, we integrated into the pipeline. The framework evaluates Logistic Regression, Random Forest, Gradient Boosting, AdaBoost, and Support Vector Machines, utilizing fairness metrics such as True Positive Rates, False Positive Rates, and statistical uniformity across demographics. Results demonstrate that Gradient Boosting achieved the best performance, with CIBIL score emerging as the dominant predictive factor (86.8% feature importance), followed by loan term (9.7%) and loan amount (1.7%), while demographic characteristics showed minimal influence. Fairness analysis across education levels revealed approval rates of 34.81% for graduates versus 39.20% for non-graduates, though statistical testing (p=0.2086) indicated no significant bias. Similarly, employment status showed minimal disparate impact with only 0.56% difference in approval rates between self-employed and traditionally employed applicants (p=0.9221). The study contributes an analytical framework that shows how credit-relevant factors can drive lending decisions without introducing demographic bias; we achieved high accuracy (>97%) while maintaining fairness across protected groups.

Abstract: The inability of the traditional mobility aids to provide safe navigation, hazard and real-time communication causes significant challenges to the blind population. In this paper, I will present a smart blind stick which is equipped with IoT sensing, a vision algorithm to practice smart blindness, and a mobile navigation application to play a bigger role in increasing the levels of independence and safety. The suggested system is based on a dual-module design, including an ESP32 integrated circuit with ultrasonic sensors, fire sensor, IMU-based fall detection, GPS, SOS mechanism including emergency messages and real-time identification of hazards, and a camera-based AI vision module, which detects objects and recognizes the surroundings based on a camera and makes independent decisions based on gestures encoded in a sign language. Besides this, it has a special navigation mobile application which gives GPS routes and therefore users are able to navigate the unfamiliar environment safely. Sensor, vision model and navigation services provide information displayed in vibration feedback, audio output and automated location-based information. The proposed system will double the main drawbacks of the traditional blind sticks by integrating navigation assistance with smart sense and emergency communication, providing a fully assistive solution to enhance the mobility, situational awareness, and quality of life of visually impaired people.

Keywords: Smart Blind Stick, Assistive Technology, Internet of Things (IoT), ESP32, Navigation Assistance, Mobile Navigation Application, GPS-Based Navigation, Obstacle Detection, Computer Vision, Object Detection, Gesture Recognition, Emergency Alert System.

Abstract: In this project, we developed a hybrid modeling technique models K-Nearest Neighbors (KNN), Random Forest, and Support Vector Machine (SVM) models to estimate the severity of diabetic retinopathy using the APTOS dataset. By merging the strengths of these three algorithms, the model delivers more stable, accurate, and dependable predictions compared to individual classifiers. Such precise severity grading can support healthcare professionals by providing early warnings and timely insights, helping them plan proactive treatment for patients at risk. Beyond improving accuracy, the proposed ensemble method also reduces inconsistencies between different diagnostic systems, making it easier to integrate into existing medical workflows. This enhances overall diagnostic reliability and promotes better clinical decision-making. The study plays a vital role in applying machine learning in real-world healthcare settings, ultimately aiming to support clinicians and result that enhanced outcomes for patients with diabetic retinopathy. Diabetic retinopathy doesn't manifest as a singular entity but progresses through stages of severity. These stages, which include mild non-proliferative, moderate non-proliferative, severe non-proliferative, and proliferative DR.

Keywords: KNN, SVM model, Diabetic Retinopathy, APTOS, Healthcare professionals, Leveraging machine.

Abstract: A smarter way to manage groceries at home, the Smart Grocery Kit blends everyday convenience with connected technology. It centers on two smart containers that feel almost like a mini smart pantry. Each container is equipped with an RFID reader, weight sensor, air quality detector, temperature monitor, and an ESP32 microcontroller. When you drop a new item into a container, its RFID tag is scanned to pull up the product name, current quantity, and expiration date. This data is instantly shared to both a Telegram bot and a web dashboard, so you can check what you have and when you’ll need to restock—without digging through drawers.The Telegram bot acts as your friendly reminder assistant. It nudges you when items are nearing their expiration, when you’re running low, or if something looks like it might spoil soon. The weight sensors keep a running tally of how much of each item is left, while the air quality detector can pick up early signs of spoilage. Temperature monitoring helps ensure foods stay in their ideal conditions, so you’re less likely to waste them.On the web side, a clean dashboard presents nutrition details alongside your inventory. You can see calorie counts and other essential facts for all stored items, helping you plan meals with health in mind. To keep you in the loop at a glance, an LCD display on each container shows live information—temperature, item name, and expiration date—so you always know what’s going on, even without checking your phone.

Keywords: IoT, ESP32, Smart Grocery, Load Cell, DHT11, MQ3 Gas Sensor, Nutrition Recommendation, Automation, Remote Monitoring, Smart Home.

Abstract: Road traffic accidents represent a critical global public health challenge, claiming millions of lives annually. A significant factor contributing to fatalities is the delay between the occurrence of an accident and emergency medical response. This paper presents AcciRescue, an intelligent real-time accident detection and alert system designed to minimise response time and improve survival rates. The system integrates NEO-6M GPS and SIM800L GPS modules with an accelerometer and gyroscope sensor connected to an Arduino Nano microcontroller to continuously monitor vehicle movement and detect collision events. Upon detecting sudden velocity changes indicative of an accident, AcciRescue automatically transmits the precise incident location to emergency contacts and nearby hospitals. The proposed system offers a cost-effective, scalable solution that can be seamlessly integrated into vehicles or smartphones, addressing the critical need for rapid emergency response in traffic accidents.

Keywords: Accident detection system, real-time monitoring, embedded systems, Arduino Nano, accelerometer, gyroscope, NEO-6M GPS module, SIM800L GSM module, emergency response, collision detection, automated alert system, traffic accident management.

Abstract: This study investigates machine learning approaches for predicting COVID-19 hospitalization rates in San Francisco, utilizing public datasets from DataSF encompassing testing metrics, deaths, and demographics from March 2020 to April 2024. The primary objective is to accurately predict daily patient counts in both Intensive Care Units and Medical-Surgical units through two distinct modeling tasks: point regression and long-horizon forecasting. For the point regression task, features were engineered from aggregated daily statistics, including lagged death counts and race-disaggregated testing data. A comparative analysis of five regressors models- K-Nearest Neighbors (KNN), Decision Tree, Linear Support Vector Machine, Non-linear Support Vector Machine, and Multi-layered Perceptron- was conducted using k-fold cross validation. Preliminary results indicate that the K-Nearest Neighbors regressor significantly outperformed other models, achieving high R² scores of 0.97 for ICU and 0.98 for Med/Surg patient predictions, demonstrating its effectiveness in capturing complex, non-linear relationships within the temporal data. For multi-horizon forecasting, Long Short-Term Memory and Gated Recurrent Unit models were trained on 120 days of data to predict 120 days in the future. Though with some deviation from the true noise of the output, these models successfully capture broader trends, indicating that COVID-19 hospitalization rates are predictable, to a degree. Overall, this research demonstrates the high efficacy of KNN for point-in-time predictions and establishes a promising baseline for deep learning-based long-term forecasting of COVID-19.

Abstract: Communicating effectively becomes extremely difficult for individuals with speech and hearing impairments, especially as the majority of people are not familiar with sign language. Existing assistive solutions often fall short as they are costly, influenced by environmental conditions, or lack real- time confirmation for the user. To overcome these limitations, the proposed IoT-based Smart Support Glove translates hand gestures into readable text and audible speech instantly. The glove integrates flex sensors and an MPU6050 module to accurately capture finger movements and hand orientation. An Arduino Nano processes these signals and identifies the performed gesture, which is then transmitted to a server through an ESP8266 Wi-Fi module. The server-side algorithm converts the encoded gesture data into text or speech within a mobile application. Additionally, a vibration motor provides haptic feedback, assuring the user whether their purpose has been delivered. Designed to be lightweight, affordable, and scalable, this wearable system offers a practical assistive tool that enhances communication, accessibility, and social inclusion.

Keywords: IoT (Internet of Things), Assistive Technology, Gesture Recognition ,Smart Glove / Wearable Device ,Flex Sensor ,MPU6050 (Accelerometer/Gyroscope) ,Arduino Nano , ESP8266 Wi-Fi Module ,Text-to-Speech (TTS) ,Haptic Feedback/ Vibration Motor / Buzzer ,Wireless Communication ,Speech and Hearing Impairment ,Embedded System.

Abstract: An AI-powered system called the Travel-Bot Planner is developed to enhance and simplify the travel planning experience for users by combining intelligent route analysis, destination discovery, and automated itinerary generation. The platform provides an interactive, adaptive, and scalable travel-assistance environment by utilizing advanced technologies such as Large Language Models (LLMs) for natural language understanding, Retrieval-Augmented Generation (RAG) for real-time information retrieval, and geospatial services for accurate routing and map-based visualization. Users are guided through AI-driven conversations where their queries, preferences, and destinations are processed using NLP and vector-based search to deliver personalized recommendations, travel timelines, and route-specific popular attractions. The system enables travellers to receive structured itineraries, budget estimates, and booking simulations instantly, reducing manual effort and ensuring informed decision-making powered by intelligent automation.

Keywords: Travel Assistance, Large Language Models (LLM), Retrieval-Augmented Generation, Geospatial Intelligence.

Abstract: High-speed arithmetic units are a critical requirement in modern digital signal processing systems and general-purpose processors. Among these units, multipliers and multiplier-and-accumulator (MAC) blocks significantly influence overall system performance. This paper presents a novel MAC architecture based on the Radix-4 Modified Booth multiplication algorithm, implemented on a Xilinx FPGA platform. The proposed design integrates multiplication and accumulation operations using an efficient hybrid adder structure, resulting in improved computational speed and reduced hardware complexity. The Modified Booth encoding technique minimizes the number of generated partial products by approximately half compared to conventional multiplication methods, thereby enhancing processing efficiency. By optimizing partial product generation and accumulation, the proposed architecture achieves faster arithmetic operations, making it suitable for high-performance DSP applications.

Keywords: Radix-4 Booth Multiplier, Multiplier-Accumulator (MAC), Modified Booth Encoding, FPGA Implementation, Digital Signal Processing, Hybrid Adder.

Abstract: Agriculture remains the major occupation and source of livelihood for a large section of the population worldwide, especially in developing countries like India. However the sector faces challenges related to weather uncertainty, resource limitation, and lack of scientific decision-making tools. Most farmers usually decide which crop to grow based on guesswork or general government advice, but this information does not always match the exact conditions of their farmland, such as soil moisture, rainfall, and temperature, which can lead to low crop yield and poor income. To overcome these problems, this study presents a simple and low-cost IoT-based crop recommendation system that uses an ESP8266 microcontroller along with sensors like DHT11 for temperature and humidity, a soil moisture sensor, and a rainfall sensor to collect real-time data from the field and suggest suitable crops. In addition, a PIR sensor is used to detect the movement of wild animals or unauthorized people, helping to protect the crops and improve overall farm safety. Collected data processing is achieved by using a rule- based algorithm, extendable to machine learning models for recommending suitable crops and triggering alerts via a Flask- based web interface. This paper covers the system architecture, hardware-software integration, and experimental validation of the introduced prototype, discussing its potential to enhance agricultural productivity and sustainability.

Keywords: Internet of Things (IoT), Precision Agriculture, ESP8266, Crop Recommendation, Intrusion Detection, Smart Farming, Flask Framework.

Abstract: Locked-In Syndrome (LIS) is an extremely serious neurological disorder that results in almost complete loss of voluntary control over muscles while preserving intact mental faculties. However, communication capabilities are significantly hampered, impacting independence, mental health, and quality of life for patients with LIS. As part of this paper, I am proposing the design of an efficient, low-cost communication support system called Gaze Connect specifically for patients with LIS employing eye-tracking gaze and blink detection methods. Also, the paper will incorporate the use of artificial intelligence with computer vision libraries for eye and blink analysis communication support with text and symbol keyboard support, voice support with text-to-speech conversion capabilities, home automation support with Internet of Things technology integration, and an added module for task-scheduling support for healthcare professionals caring for patients with LIS.

Keywords: Locked-in Syndrome, Eye Gaze Detection, Blink Recognition, Virtual Keyboard, Text to Speech, IoT Automation, Assistive Technology.

Abstract: Financial markets generate large volumes of fast- moving data and require decisions to be taken in very short intervals. Human traders alone struggle to monitor all signals and react consistently without emotional bias. To address this challenge, this work presents an AI driven trading bot that combines Machine Learning (ML) for price forecasting with Reinforcement Learning (RL) for action selection. The system uses technical indicators, a Long Short-Term Memory (LSTM) network for short-term prediction and a Deep Q-Network (DQN) agent to learn profitable buy, sell and hold policies. The complete solution is deployed as a web application that provides real- time charts, portfolio analytics, sentiment summaries and AI- generated trading signals. Experimental evaluation indicates promising accuracy, low-latency inference and improved profit consistency when compared with simple rule-based strategies.

Keywords: Stock Market, Machine Learning, Reinforcement Learning, LSTM, DQN, Trading Bot, Financial Analytics.

Abstract: The analysis of count data has become increasingly important in financial and managerial research, particularly when information is collected in the form of frequencies or event counts. Conventional linear regression models are often unsuitable for such data due to their discrete and non-negative nature. Poisson regression provides an effective alternative by modelling count data within a probabilistic framework. This study applies Poisson regression analysis using statistical and data science tools to examine financial statement related count data collected from Konigtronics Private Limited. Primary data were obtained through a structured questionnaire survey from 63 respondents and supported by secondary data from company records. Since the responses represent frequency-based observations, Poisson regression was employed and parameters were estimated using maximum likelihood estimation. Descriptive statistics and correlation analysis were also used to support the findings. The results indicate that identifying relationships among financial variables, particularly cash flow patterns and reporting practices, improves the quality and reliability of financial statements. The study concludes that Poisson regression is a suitable and effective tool for analysing count data and supporting informed financial decision making.

Keywords: Poisson Regression, count data analysis, financial statements, maximum likelihood estimation, cash flow analysis, financial reporting, statistical modelling, managerial decision making.

Abstract: The scarcity of water and energy continues to be a major obstacle for sustainable agriculture, necessitating integrated and cost-effective solutions. This project introduces a hybrid framework that incorporates rainwater harvesting, solar energy utilisation, and intelligent automation to optimise resource use in agricultural fields. The cube-structured solar panel system is designed to maximise space efficiency by simulta- neously harvesting rainwater and generating renewable energy. The collected water is filtered and stored in designated tanks, while additional runoff is diverted for future use. A machine learning model predicts water quality in real-time to ensure its safe and effective utilisation by analysing parameters such as pH, turbidity, conductivity, and microbial content. These predictions power an Arduino-based control system that automatically routes water to the field for irrigation or heating for domestic and sterilisation purposes using solar energy. By integrating harvest- ing, prediction and automation, the system increases agricultural sustainability and reduces dependence on conventional resources, and also promises to be highly scalable for both rural and urban applications.

Keywords: Rainwater Harvesting, Solar Energy, Machine Learning, Water Quality Prediction, IoT, Arduino, Sustainable Agriculture

Abstract: This project develops a comprehensive vehicle predictive maintenance system that leverages IoT sensors, machine learning, and cloud computing to monitor critical vehicle components and predict potential failures before they occur. The system addresses the growing need for proactive maintenance strategies in the automotive industry by monitoring real-time parameters such as brake pad pressure, gearbox usage patterns, clutch stress cycles, and environmental humidity levels using dedicated sensors connected to an ESP32/Arduino microcontroller. Data is transmitted to cloud storage via Wi-Fi for advanced analytics and machine learning-based failure prediction. The system features a web dashboard for real-time monitoring, historical data visualization, and maintenance scheduling recommendations. Multiple alert mechanisms including local buzzers, LED/OLED displays, SMS, and email notifications ensure timely maintenance interventions.

Keywords: IoT, ESP32, Brake and Pressure Sensor, Temperature sensor, Vibration sensor, Gearbox Monitoring, Clutch Health, Firebase Cloud, Machine Learning, Predictive Maintenance.

Abstract: Financial institutions increasingly rely on cloud computing to support mission-critical workloads such as real-time payments, trade execution, regulatory reporting, and fraud analytics. While public cloud platforms offer elasticity and advanced managed services, exclusive dependence on a single provider introduces concentration risks, vendor lock-in, and exposure to regional outages, whereas private cloud environments alone can limit scalability and innovation. Mission-critical financial applications must also meet stringent requirements for fault tolerance, data protection, uninterrupted availability, and compliance with global privacy regulations such as GDPR. This paper provides a comprehensive examination of why a hybrid cloud strategy—integrating private and public cloud capabilities—is essential for achieving resilience, high availability, data sovereignty, and regulatory alignment in the financial sector. Through analysis of architectural patterns, resiliency engineering principles, operational considerations, and emerging industry practices, the paper demonstrates that a well-governed hybrid cloud model offers a balanced and robust approach for managing performance, security, and risk across modern large-scale financial systems.

Keywords: Hybrid Cloud strategy, Financial Systems, GDPR, Resiliency, Fault Tolerance, Cloud Governance, Cloud architecture, mission critical software systems

Abstract: MatdaanX is a decentralized hybrid voting architecture designed to deliver secure, transparent, and reliable election processes through the integration of blockchain technology, biometrics, and IoT-based offline modules. MatdaanX is a secure and transparent electronic voting platform developed to make the election process more dependable and user friendly. The online module authenticates voters through multiple layers of verification, including facial recognition based on MTCNN, OTP verification, and digital signatures. This ensures that each vote is genuine and securely encrypted before being submitted. In regions with poor or unstable internet connectivity, the system offers an offline setup that operates using an Arduino Uno microcontroller integrated with a fingerprint sensor, keypad, and LCD display. This allows voters to cast their ballots effortlessly without relying on continuous network access, ensuring inclusive participation in all areas. Both online and offline votes in MatdaanX are securely recorded in a private blockchain network that employs SHA-256 hashing along with a Proof- of-Authority(POA) consensus model. This approach guarantees data integrity, transparency, and resistance to any form of tampering or unauthorized modification By integrating biometric verification, blockchain-based recordkeeping, and dual online-offline connectivity, MatdaanX effectively tackles major electoral challenges such as voter fraud, duplicate entries, ballot manipulation, limited accessibility, and delays in result compilation. Overall, it offers a transparent, verifiable, and inclusive framework for conducting elections in a modern democratic environment.

Keywords: Blockchain, Biometric Authentication, IoT, Secure E-Voting, Face Recognition, Fingerprint Scanning, Decentralized Systems.

Abstract: Efficient management of doctor appointments and patient interactions is a major challenge in healthcare systems, particularly in institutes relying on traditional coordination and record-keeping. Conventional hospital administration methods are error-prone, time-consuming, and without real-time communication between doctors and patients. This paper illustrates the concept and design of a web-based Doctor Panel Management System using the MERN stack, with a list of features such as automation of appointment scheduling, payment processing, and communication through an integrated AI chatbot. It includes a responsive admin panel designed with React and Tailwind CSS, using the Context API for optimized state management. A RESTful API layer using Node.js/Express will securely interact with MongoDB, while the Razorpay integration offers a reliable and seamless digital payment mechanism. The AI chatbot enhances patient engagement with automation for appointment status tracking and queries. Experimental results show a reduction in manual effort by 40% and increased reliability within appointment scheduling. This research focuses on the potential of full-stack automation combined with AI in increasing accessibility to healthcare and operational efficiency. Future work may look into integrating predictive analytics and voice-based AI assistants for advanced healthcare automation.

Keywords: Doctor Panel System, MERN Stack, Healthcare Automation, AI Chatbot, Razorpay, RESTful API, Context API

Abstract: Wireless roadway charging is rapidly becoming a promising alternative to traditional plug-in and stationary EV charging techniques, which frequently suffer from long wait times and limited convenience. In this project, we present a solar-powered dynamic charging system capable of supplying energy to electric cars when they are moving. The system uses photovoltaic panels placed along or integrated into the roadway to harvest solar energy, which is then used to energize inductive transmitter coils embedded beneath the road surface. As an EV drives over these coils, its onboard receiver coil captures the transmitted energy, enabling continuous, interruption-free charging and reducing concerns related to battery range. Since solar energy output varies with weather conditions, the system incorporates a Long Short-Term Memory (LSTM) deep learning model to accurately forecast factors such as solar irradiance, temperature, and cloud cover. These forecasts aid in estimating power availability in real time and guarantee a steady and dependable charging process. The method is appropriate for future smart and sustainable transportation networks because experimental testing shows stable wireless power transfer, precise weather forecasting with a mean absolute error of about 24, and effective integration of all system modules.

Keywords: Dynamic Wireless Charging, Solar Energy, Electric Vehicles, LSTM Weather Forecasting, Inductive Power Transfer, Embedded Systems

Abstract: Conventional computer input devices impose physical constraints and accessibility barriers where traditional keyboards and mice provide precise control but require direct surface contact and fine motor coordination. This research presents HandPilot, an integrated wearable computing framework combining multi-sensor fusion with supervised machine learning for intuitive gesture-based human-computer interaction. The system deploys Arduino-based smart glove modules with MPU6050 6-axis inertial measurement unit (3-axis accelerometer and 3-axis gyroscope), resistive flex sensor for finger bend detection, and three tactile push buttons to capture spatially oriented hand movements and discrete click operations. Structured data packets containing accelerometer readings (Ax, Ay, Az), gyroscope measurements (Gx, Gy, Gz), flex sensor values, and button states are transmitted wirelessly via HC-05 Bluetooth module at 9600 baud rate to a receiver Arduino connected to host computer through USB serial interface. A comprehensive labeled dataset mapping ten gesture classes—right, left, up, down, zoom in, zoom out, drag, left-click, right-click, and double-click—to corresponding sensor feature vectors enables training of multiple supervised classification algorithms including K-Nearest Neighbors (KNN), Support Vector Machines (SVM) with RBF kernel, Decision Trees with Gini impurity splitting, and Random Forest ensemble methods. The trained Random Forest model integrates into a Python-based real-time control engine using PySerial for serial communication and data parsing, combined with PyAutoGUI for cross-platform mouse action execution and cursor control. Experimental validation demonstrates superior performance, achieving 94% classification accuracy with the Random Forest classifier, representing 30% improvement over conventional threshold-based gesture recognition approaches, with end-to-end system latency of 80-120ms and 99.7% data transmission reliability during 24-hour continuous operation testing.

Keywords: Gesture Recognition, Wearable Computing, Machine Learning, Human-Computer Interaction, MPU6050, Bluetooth Communication, Assistive Technology

Abstract: Growing cities mean more apartments and offices sit empty during work hours or vacations, making break-ins a real problem. Sure, CCTV cameras and motion alarms exist everywhere now, but honestly? They mostly just record stuff passively or beep after someone’s already inside. You end up with terabytes of useless footage nobody watches, plus everyone feels weird about cameras recording them 24/7. There’s got to be a smarter approach. Our solution started simple: what if your floor mat could think? We built this thing using cheap IoT parts from Amazon and AliExpress. Basically, there’s a pressure sensor hidden in a regular-looking floor mat at the entrance. Step on it, and a Force Sensitive Resistor notices. But we didn’t stop there—there’s also an infrared motion detector watching the same spot. Why both? Because my cat weighs enough to trigger pressure sensors, and shadows can fool motion detectors. The Arduino Uno microcontroller waits until BOTH sensors agree someone’s actually there before doing anything. Only then does an ESP32-CAM module wake up and snap pictures. Those images get crunched through OpenCV running Local Binary Pattern Histogram face matching—we fed it photos of everyone who lives there plus some random faces for testing. The ESP8266 NodeMCU chip grabs the ”recognized” or ”stranger alert” result and pushes it through n8n workflows straight to your Telegram app. Your phone buzzes within seconds. We ran this through its paces with 200 staged intrusions. Different times of day, various lighting situations, people wearing hats, you name it. Got 94.2% correct identifications, and the whole chain from footstep to phone notification averaged 2.8 seconds. Maybe five or six false alarms total across all those tests. The best part? Camera only runs when someone’s actually at the door, so no creepy always-on recording. Works great for regular houses, dorm rooms, small startups—anywhere you can stick a mat by the door. Index Terms: IoT Security, Smart Home Systems, ESP32-CAM, Face Recognition, Intru- sion Detection, n8n Automation, Telegram Bot.

Abstract: The rapid growth of vehicular traffic on Indian highways has exposed major limitations in conventional toll collection systems, including long queues, manual cash handling, and service disruptions. This paper presents a smart automated toll gate system that integrates RFID-based toll collection with vehicle monitoring and driver assistance features to improve traffic flow and user experience. The proposed system enables automated toll deduction while allowing controlled vehicle passage even in cases of insufficient balance, thereby preventing congestion. An in-vehicle interface provides real-time balance notifications and service request options for fuel, mechanical help, or towing. Additionally, an idle monitoring mechanism detects stalled vehicles and initiates polite alerts to maintain lane efficiency. The system is designed to be modular, low-cost, and compatible with existing FASTag infrastructure, making it suitable for practical deployment in real-world toll plazas.

Keywords: RFID, Automated Toll Collection, Intelligent Transportation Systems, Vehicle Monitoring, Driver Assistance System, FASTag, Embedded Systems, Smart Toll Plaza

Abstract: This paper presents the design and implementation of a digital Proportional-Integral-Derivative (PID) controller using Field Programmable Gate Array (FPGA) for precise speed control of a DC motor. The proposed PID controller enhances motor speed stability by continuously generating the control signal in real time using the difference between the reference speed and the measured motor output. The digital PID controller demonstrates efficient motor speed regulation, reduces the steady-state error, and fast response time. This FPGA-based implementation offers flexibility, scalability, and high-speed operation, making it suitable for various industrial automation and embedded control applications.

Keywords: Digital pid controller, FPGA implementation, dc motor control, Verilog hdl.

Abstract: Coffee farming faces challenges such as crop diseases, improper fertilizer usage, climate variability, and limited access to expert guidance. Traditional practices rely on manual observation, leading to delayed disease detection and uncertain yield outcomes. This paper presents SmartCrop-Coffee, an AI-based decision support system for precision coffee farming. The framework integrates deep learning and machine learning to enable coffee leaf disease detection, fertilizer recommendation, coffee variety selection, and yield prediction. A Convolutional Neural Network (CNN) classifies leaf diseases, while Random Forest models support fertilizer, yield, and variety prediction using soil and crop data. Real-time weather data further enhances decision accuracy, supporting sustainable and data-driven coffee agriculture.

Keywords: Precision Agriculture, Coffee Leaf Disease Detection, Machine Learning, CNN, Yield Prediction.

Abstract: Sleep disorders and poor sleep quality have become common health concerns due to increased screen exposure, academic stress, irregular work schedules, and modern lifestyle habits. Inadequate sleep negatively affects physical health, mental well-being, concentration, and productivity. Conventional sleep aids such as basic eye masks or medication-based solutions provide limited effectiveness and may introduce discomfort or side effects. This paper presents Marphee: The Smart Sleeping Mask, a wearable and non-invasive sleep assistance system designed to enhance sleep quality through intelligent sensing and comfort-oriented features. The proposed system focuses on reducing external disturbances, promoting relaxation, and supporting natural sleep cycles without pharmaceutical intervention. By integrating smart wearable technology with user-friendly design, Marphee offers a safe, portable, and effective solution for individuals experiencing sleep-related difficulties. The system aims to improve overall sleep efficiency, mental relaxation, and daily performance, making it suitable for students, professionals, and individuals affected by sleep irregularities.

Keywords: Smart Wearable, Sleep Monitoring, Sleep Quality Improvement, Smart Sleeping Mask, Health Technology

Abstract: S-band (2-4 GHz) antennas play a critical role in modern wireless systems such as satellite communication, Wi-Fi, Bluetooth, and mobile networks. Microstrip Patch Antennas (MSPAs) are widely preferred for these applications due to their compact size, low profile, and ease of fabrication. However, conventional MSPA designs often exhibit narrow bandwidth, low gain, and reduced radiation efficiency because of strong nonlinear coupling among design parameters. This paper presents an Artificial Intelligence (AI) based optimization approach using Fuzzy Logic (FL) to enhance the performance of S-band MSPA. A Fuzzy Inference System (FIS) is developed to intelligently tune critical antenna parameters including patch dimensions and feed location. The paper proposes the design is implemented in MATLAB R2013a [8] and evaluated at a target frequency of 2.6 GHz. Simulation results demonstrate considerable improvement in antenna performance, achieving enhanced bandwidth, improved gain, and acceptable impedance matching. Comparative analysis confirms that the fuzzy logic optimized MSPA outperforms the conventionally designed antenna in terms of bandwidth, compactness, and radiation characteristics, making it suitable for modern S-band wireless communication applications in the current scenario.

Keywords: S-band, Microstrip Patch Antenna, Fuzzy Logic, Artificial Intelligence, Antenna Optimization, Bandwidth, Gain.

Abstract: The Smart Mining Helmet is an innovative Internet of Things (IoT)–based system designed to improve safety and real-time monitoring in hazardous underground environments. It integrates a variety of sensing technologies to detect dangerous gases, track miner health, and provide continuous environmental awareness with minimal manual effort. The system includes a durable helmet equipped with gas sensors, temperature and humidity modules, a heartbeat sensor, and an ESP8266 microcontroller. When a miner enters a work zone, the sensors measure gas levels, thermal conditions, and vital signs to record critical safety information. This data is instantly transmitted to a cloud platform and a monitoring interface for immediate access. Alerts are generated when gas concentrations exceed safe limits, when temperature or humidity becomes unstable, or when abnormal heartbeat values are detected. The gas sensors monitor early traces of harmful emissions, while the environmental modules help maintain awareness of unsafe atmospheric shifts. The interface displays live data such as gas intensity, miner identification, temperature, and heart rate. Overall, the Smart Mining Helmet enhances worker protection, reduces risk, and supports rapid emergency response by combining automation, sensing, and real-time analysis into a single IoT-driven safety solution.

Abstract: The credibility of democratic systems relies heavily on the integrity and transparency of electoral processes. Traditional voting mechanisms, including paper ballots and electronic voting machines, suffer from several drawbacks such as centralized control, susceptibility to manipulation, lack of transparency, and delayed result declaration. To address these limitations, this paper proposes a Blockchain-Based E-Voting System that leverages decentralized ledger technology to ensure secure, transparent, and tamper-proof elections. The system utilizes Ethereum-compatible blockchain networks and smart contracts written in Solidity to automate vote validation, storage, and counting. A modern frontend built using React, TypeScript, and Vite provides an intuitive voting interface, while Supabase serves as a cloud backend for authentication, election metadata, and audit logging. Wallet-based voter authentication using MetaMask ensures secure participation and prevents duplicate voting. Experimental evaluation demonstrates that the proposed system significantly improves trust, efficiency, and transparency while reducing operational overhead. The results validate blockchain as a viable solution for secure digital voting systems.

Keywords: Blockchain, E-Voting, Smart Contracts, Ethereum, Solidity, Supabase

Abstract: Rapid and reliable movement of emergency vehicles is critical for saving lives, yet conventional traffic signal systems often fail to provide timely right-of-way under congested urban conditions. This work presents a reinforcement learning (RL) based emergency vehicle prioritization framework enhanced by Vehicle-to-Everything (V2X) communication and evaluated using the SUMO traffic simulator. The proposed system enables traffic signals to dynamically adapt their phases based on real-time information exchanged between emergency vehicles, roadside units, and intersections. An RL agent is trained to minimize emergency vehicle delay while maintaining overall traffic efficiency by observing traffic density, queue lengths, and emergency vehicle proximity. V2X communication ensures early detection of approaching emergency vehicles, allowing proactive signal control rather than reactive pre-emption. Simulation results demonstrate that the proposed approach significantly reduces emergency vehicle travel time and intersection delay compared to fixed-time and conventional priority strategies, while limiting negative impacts on non-emergency traffic.

Keywords: Emergency Vehicle Prioritization, Reinforcement Learning, V2X Communication, Intelligent Traffic Signal Control, SUMO Simulation, Smart Transportation Systems

Abstract: Railway safety continues to be a critical challenge in modern transportation systems due to recurring issues such as track defects, wildlife interference, and fire-related incidents, which frequently lead to severe accidents. This paper presents an intelligent IoT-enabled autonomous robotic system designed to address these challenges through a multi-modal hazard detection approach. The proposed robot autonomously navigates railway tracks while capturing visual data for structural assessment using edge detection techniques and Hough transformation to identify cracks and defects. In parallel, the system monitors wildlife intrusion and fire hazards to prevent potential collisions and emergencies. Detected information is transmitted wirelessly via IoT infrastructure to control centers, enabling real-time monitoring and rapid response actions. Machine learning models, including YOLO for object recognition and MobileNet-SSD for wildlife detection, are utilized to enhance detection accuracy. Experimental results from field tests demonstrate reliable performance with a low false alarm rate and high detection accuracy, while maintaining a cost-effective implementation of approximately INR 5,000 per unit. By reducing dependence on manual inspections and enabling continuous surveillance, the proposed system significantly enhances railway safety and operational efficiency.

Keywords: Railway safety, Autonomous robot, IoT-based monitoring, Computer vision

Abstract: An Intrusion Detection System (IDS) is a critical security mechanism designed to monitor network traffic and system activities to identify malicious actions or policy violations. With the rapid growth of interconnected systems and the increasing sophistication of cyberattacks, traditional security solutions such as firewalls are no longer sufficient on their own. This project focuses on the design and implementation of an effective intrusion detection system that enhances network security by identifying both known and unknown attacks in real time. The proposed IDS analyzes incoming data packets and system behavior to detect abnormal patterns that may indicate unauthorized access, denial-of-service attacks, or data breaches. Machine learning techniques are employed to learn from historical data and classify activities as normal or malicious, thereby improving detection accuracy and reducing false alarms. The system continuously adapts to new attack patterns, making it suitable for dynamic network environments.

Keywords: Intrusion Detection System, Network Security, Cyber Attacks, Machine Learning, Anomaly Detection, Signature-Based Detection, Network Monitoring, Malicious Traffic, Threat Detection, Data Analysis

Abstract: In our recent years, we have faced the problem of a gap between the non-governmental organizations (NGOs) and society has become increasingly vital for addressing community-level issues such as sanitation, waste management, water leakage, and environmental degradation. However, the absence of a unified digital platform often leads to fragmented communication and inefficient issue resolution. JanMitra is a next-generation AI-powered social connection platform designed to bridge this gap by connecting societies, NGOs, and government bodies within a single ecosystem. The system leverages Artificial Intelligence (AI) for image recognition to automatically identify the nature of a reported problem—such as garbage accumulation, water leakage, or road related problem and intelligently routes the complaint to the verified and registered NGO. The platform features a mobile application for residents and society heads, a web dashboard for NGOs and government organizations, and a core backend engine that handles AI-based classification, routing, and analytics.

Keywords: Artificial Intelligence, Image Recognition, NGO Collaboration, Civic Engagement, Smart Society, Social Innovation.

Abstract: In order to guarantee that agricultural produce reaches customers in a safe, fresh, and nourishing state, the food supply chain is essential. However, issues like contamination, spoiling, nutritional deterioration, lack of transparency, and ineffective traceability plague conventional food monitoring systems. Due to the dependence on human inspections and centralized record-keeping systems, these restrictions raise health risks, cause monetary harm, and erode customer confidence. In order to guarantee food safety and quality along the whole supply chain, the Food Supply Chain Health Tracker is a clever, technologically advanced solution that combines Blockchain, AI, and ML. While ML models forecast nutritional degradation based on storage and climatic conditions, AI-based computer vision techniques are utilized to identify contamination and spoiling in crops. Blockchain technology ensures transparency and end-to-end traceability from farm to consumer by providing a decentralized, unchangeable ledger for all supply chain transactions. Through a web-based platform, the system provides farmers, distributors, retailers, and customers with role-based access. While stakeholders receive real-time information on food quality and handling circumstances, consumers can use QR code scanning to confirm product quality and origin. The suggested solution improves food safety, lowers post-harvest losses, increases responsibility, and fosters confidence in the food supply ecosystem by fusing secure traceability with predictive analytics.

Abstract: Elections are a critical component of democratic governance, yet traditional voting systems continue to face significant challenges such as voter impersonation, duplicate voting, manual verification errors, and lack of transparency. These issues undermine public trust and election integrity. To overcome these limitations, this paper presents an offline Smart Voting System through Face Recognition that employs deep learning–based biometric authentication for secure and reliable voter verification. The proposed system performs voter registration using Aadhaar as a unique identifier and captures multiple facial samples through a webcam. Facial embeddings are extracted using a pre-trained deep learning model and stored locally using serialized pickle files. During the voting phase, real-time facial recognition is performed using cosine similarity, enhanced by temporal smoothing and face tracking to improve accuracy and stability. Votes are recorded securely in CSV format, ensuring transparency and preventing duplicate voting. Experimental evaluation demonstrates high recognition accuracy, low false acceptance rates, and efficient real-time performance, making the system suitable for secure offline voting environments.

Keywords: Smart Voting System, Face Recognition, Deep Learning, Biometric Authentication, Aadhaar Verification, Election Security.

Abstract: Stress has become a major concern among students in higher education, particularly due to academic demands and socio-economic pressures. The present study aimed to examine the overall level of stress and types of stress among scholarship holder and non-scholarship holder students in higher education institutions of the Marathwada region of Maharashtra. A descriptive, cross-sectional research design was adopted for the study. The sample consisted of undergraduate and postgraduate students selected through a random sampling technique. Data were collected using self-reported questionnaires, including a demographic information schedule and a stress assessment questionnaire measuring overall stress levels (mild, moderate, and severe) and types of stress (acute stress, chronic stress, and eustress). The results revealed that 48.82% of scholarship holder students experienced mild stress, 28.66% moderate stress, and 22.34% severe stress, whereas 54.66% of non-scholarship holder students reported mild stress, 27.33% moderate stress, and 20.50% severe stress. With regard to types of stress, 52.12% of scholarship holder students reported acute stress, 32.45% chronic stress, and 16.66% eustress. In contrast, non-scholarship holder students showed lower levels of acute (40.33%) and chronic stress (28.30%) but a higher level of eustress (31.56%). The findings indicate that scholarship holder students experience comparatively higher levels of distress, possibly due to academic performance pressure and financial dependency. The study highlights the need for targeted stress management interventions and institutional support systems to promote mental well-being among higher education students.

Keywords: Academic Stress, Scholarship Holder Students, Non-Scholarship Holder Students, Higher Education, Mental Health, Marathwada Region

Abstract: The present study aimed to examine psychological problems with respect to stress among swimmers and non-swimmers in the age group of 24–30 years. A total of 185 participants were selected for the study, including 69 swimmers and 116 non-swimmers. Stress levels were assessed using a standardized psychological stress scale. Descriptive statistics such as mean scores and standard deviations were calculated, and an independent samples t-test was employed to determine the significance of differences between the groups. The results indicated that non-swimmers (M = 22.78, SD = 4.87) exhibited significantly higher stress levels compared to swimmers (M = 19.23, SD = 4.32). The obtained t-value (t = 3.67) was statistically significant at the 0.05 level. The findings suggest that regular participation in swimming may play an important role in reducing perceived stress among young adults. The study highlights the psychological benefits of swimming and supports the inclusion of physical activity as a strategy for stress management.

Keywords: Stress, Psychological Problems, Swimmers, Non-Swimmers, Physical Activity, Mental Health.

Abstract: Modern urban waste management systems tend to make use of artificial intelligence (AI) and big data technologies through the collection of multimodal Internet of Things (IoT) data to manage operational inefficiencies and unsustainability’s. Scope of a broad-stroke synthesis of AI and big data features and applications for smart urban waste management, alongside the relevance of established AI and smart city concepts, with conclusions that point to critical pathways for application and research. Actionable real-world conclusions naturally arise from a deeper understanding of a broad stroke thinking on smart waste as well as the interrelation between AI capabilities and urban waste management drivers. A smart waste management system encompasses the entire urban waste lifecycle, from generation and collection to recycling and reprocessing, focusing on the generation, collection, sorting, and recycling steps; and processes driven by data fusion and artificial intelligence. Urban waste systems logically collect heterogeneous data to inform operation. The potential of modern smart waste concepts rests on Internet of Things (IoT) and data-driven technologies applied to waste systems. The overwhelming amount of novel sensing devices, capable of gathering information about waste fill levels and additional smartness features provide the ability to create real-time fill level forecasts. Apart from the sensing on bins, smart containers, capable of providing additional information (e.g., temperature, smoke) have also been deployed. Twofold analysis improves fill level forecasting through anomalies detection and resolution. All those novelties create a need for a transversal analysis of all the innovations, elements, and data-enabled technologies proposed through a smart waste concept.

Keywords: Smart Urban Waste Management, AI-Driven Waste Systems, Big Data In Waste Management, IoT-Enabled Waste Collection, Multimodal Waste Sensing, Waste Lifecycle Analytics, Real-Time Fill Level Forecasting, Smart Waste Containers, Waste Anomaly Detection, Data Fusion For Waste Systems, Urban Sustainability Technologies, Intelligent Waste Collection Optimization, Smart City Waste Solutions, Waste Sorting And Recycling Analytics, Sensor-Based Waste Monitoring, Predictive Waste Management, Heterogeneous Urban Data Integration, Operational Efficiency In Waste Systems, AI Applications In Smart Cities, Data-Driven Urban Sustainability.

Abstract: AI-based cloud systems are expected to have a positive effect in enabling efficient automated processing of standard legal documents. Legal technology (LegalTech) tools aim to increase efficiency in automated legal services such as e-discovery and legal-billing review by classifying, extracting, comparing, and summarizing information. These document-specific tasks rely on supervised-computing models that require large-scale datasets for training and performance evaluation. Cloud-based services based on multi-task and multi-lingual-large-pretrained transformer models are proposed for supporting the automation of common LegalTech tasks, including contract analysis, abbreviation, e-discovery, and litigation support. LegalTech service providers usually offer platform-as-a-service or software-as-a-service solutions to support the e-discovery process—all of which require compliance with legal and ethical regulations. Therefore, deployment of AI services must guarantee not only satisfactory accuracy and performance metrics but also issues such as data governance, bias determination, mitigation procedures, accountability assignment, and ethical compliance of usage. Concentrating on the architectures that provide these services, the availability of the AI models for Cloud APIs covering the required tasks is paramount.

Keywords: AI-Based LegalTech Systems, Cloud Legal Document Automation, Automated Legal Services, E-Discovery Analytics, Legal Billing Review Automation, Contract Analysis AI, Legal Text Classification, Information Extraction In Law, Multi-Task Transformer Models, Multilingual Legal AI, Cloud-Native Legal Platforms, LegalTech SaaS And PaaS, Supervised Learning For Legal Data, Legal Data Governance, Bias Detection And Mitigation, Ethical AI In Legal Services, Regulatory Compliance In LegalTech, Accountability In AI Systems, Cloud APIs For Legal AI, Scalable Legal Analytics.

Abstract: Data-driven models leveraging artificial intelligence (AI) and big data offer the potential for earlier detection of emerging disease outbreaks over traditional approaches. They operate with real-time visibility, can explore a broad threat landscape, and submit signals with varying reliability. Such capabilities can address a perennial challenge in infectious disease surveillance: signal generation that is timely enough to meaningfully inform response efforts. Yet despite this apparent potential, these models remain largely unexploited in public health. A candidate framework for operationalization and two case studies demonstrate the pathway: COVID-19 incidence time series models employing social media signals and long-range influenza signals for a major city in a resource-rich country-making timely signals available to public health decision-making. AI- and big-data-enabled outbreak models present an alternative detection approach that shifts traditional epidemiological assumptions. Early warnings derived from these models have distinct characteristics. Alerts can emerge at shorter lead times, multiplexed requests—demanding different signals responding to distinct factors—can be launched simultaneously, and AI-based models can harness digital exhaust, unfiltered datasets generated as by-products of everyday human activity. Such a vast volume of high-frequency data could thus enable early warning systems to submit multiple signals with different reliability scores at little additional operational overhead.

Keywords: AI-Driven Disease Surveillance, Big Data Epidemiology, Early Outbreak Detection, Real-Time Public Health Analytics, Infectious Disease Forecasting, Digital Disease Signals, Social Media Epidemiology, AI-Based Early Warning Systems, Public Health Decision Support, Emerging Disease Monitoring, High-Frequency Health Data, Signal Generation And Validation, Multiplexed Surveillance Signals, Pandemic Preparedness Analytics, Influenza Forecasting Models, COVID-19 Time Series Analysis, Digital Exhaust Data, Risk Scoring For Outbreaks, Operational Public Health AI, Next-Generation Epidemiological Models.

Abstract: Elderly women living in urban slums represent one of the most vulnerable population groups in India due to the intersection of ageing, gender inequality, and socio-economic deprivation. This meta-analysis aims to synthesize existing evidence on the health conditions of elderly women residing in slum areas, focusing on physical morbidity, mental health, and socio-economic determinants. A systematic search of databases including PubMed, Scopus, and Google Scholar was conducted for studies published between 2000 and 2024. A total of 18 studies met the inclusion criteria, covering approximately 12,500 elderly women aged 60 years and above residing in urban slums. The pooled prevalence of multimorbidity among elderly women was found to be extremely high, ranging from 62% to 95%. Common conditions included hypertension, musculoskeletal disorders, dental problems, and diabetes. Mental health issues were also significant, with depression prevalence ranging from 31% to 48%, often associated with financial dependency, social isolation, and lack of family support. Poor health literacy, inadequate access to healthcare services, and unfavorable living conditions such as overcrowding and poor sanitation further exacerbate health risks. Social determinants such as widowhood, illiteracy, and economic dependency significantly influence health outcomes among elderly women in slums. The findings highlight a critical need for targeted interventions focusing on gender-sensitive healthcare, improved access to services, and social support mechanisms. Strengthening community-based health programs and improving health literacy are essential to enhance the quality of life of elderly women in slum settings.

Keywords: Elderly Women, Slums, India, Multimorbidity, Depression, Health Inequality, Meta-analysis