VOLUME 15, ISSUE 2, FEBRUARY 2026

Abstract: Urban traffic congestion poses a major, worldwide problem, resulting in substantial financial losses, increased air pollution, and a decline in the overall quality of life. Current monitoring methods, such as embedded loop sensors and cameras positioned on roadsides, are often restricted in their coverage, expensive to maintain, and suffer from insufficient data in areas that lack adequate instrumentation. This study introduces a new, easily scalable strategy for detecting traffic congestion in real-time and predicting it in the short term. This method relies on analyzing the density of aggregated and anonymous mobile device location data. We exploit the widespread availability of mobile phones as ubiquitous, cost-effective sensors to gather precise spatial and temporal data about how vehicles are moving. The proposed technique involves dividing the urban area into a uniform grid, calculating a constantly changing mobile device density for each section, and combining this with estimated average vehicle speeds to produce a comprehensive Congestion Index. ( ). We tested and compared three sophisticated Artificial Intelligence (AI) models; Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Long Short-Term Memory (LSTM) networks; to classify current traffic jam levels and anticipate future conditions. The results, based on a vast, anonymous dataset from a major urban area, clearly show that the LSTM model's time-series forecasting capability is superior to the tree-based ensemble methods for short-term prediction. It achieved an impressive score of 0.94 and a minimal Mean Absolute Error (MAE) of 0.05 when predicting congestion. This method, which relies on density analysis, presents a reliable, economical, and easily scalable replacement for expensive traditional infrastructure, offering city planners and traffic managers immediate, practical information to effectively reduce traffic jams.

Keywords: Mobile location data, Traffic congestion detection, Artificial intelligence, Smart cities, Density analysis

Abstract: Recent progress in Large Language Models (LLMs) has significantly advanced natural language processing tasks such as summarization, translation, and text generation. Despite their impressive capabilities, these models frequently generate hallucinations—responses that appear fluent and convincing but lack factual correctness or logical grounding. Such behavior raises serious concerns regarding the dependability and ethical deployment of LLMs in real-world scenarios. This paper reviews and analyzes existing research on hallucinations in LLMs, focusing on their underlying causes, practical consequences, and mitigation strategies. Studies by Reddy et al. (2024) and Perković et al. (2024) investigate both internal model limitations and external influencing factors, including biased datasets, inadequate contextual understanding, and poorly structured prompts. Their findings highlight ethical and operational risks across multiple application domains. Research presented at ICALT 2024 emphasizes the dangers of hallucinated content in educational environments and proposes comparative and cross-verification techniques to preserve factual integrity. Furthermore, Sun et al. (2025) introduce a Markov Chain–based multi-agent debate framework that enhances post- generation verification through structured evidence retrieval and claim validation.

Abstract: Agriculture is essential to the economies of countries like India, but it faces challenges such as changing climate, poor soil nutrients, and outbreaks of plant diseases. To tackle these problems, this study introduces a Smart Crop and Fertilizer Recommendation System with Plant Disease Identification that uses machine learning and deep learning techniques. The web-based system gives real-time suggestions for suitable crops and fertilizers based on data about soil nutrients, temperature, humidity, pH, and rainfall. We trained and evaluated seven machine learning models: Decision Tree, Naive Bayes, Support Vector Machine (SVM), Logistic Regression, Random Forest, XGBoost, and K-Nearest Neighbours (KNN) for crop recommendations. The Random Forest model achieved the highest accuracy and was chosen as the best option. Additionally, the system generates fertilizer recommendations to fix soil nutrient shortages and boost crop yield. The system also includes plant disease identification using a Convolutional Neural Network (CNN), which analyses leaf images to classify diseases accurately. This allows for early detection and timely response. This integrated solution helps users make informed decisions, supports sustainable farming, and improves productivity and food security.

Keywords: CNN, crop recommendation, machine learning, plant disease identification, random forest

Abstract: The rapid digital transformation of healthcare has increased the need for secure, transparent, and intelligent data-management systems. HealthChainSync is an Android-based dual-role application designed to provide a tamper-proof health record ecosystem using Blockchain, SHA-256 hashing, and Firebase-backed cloud synchronization. The system enables patients to upload medical documents, diagnostic reports, and health parameters securely, while doctors can review, validate, and update records with blockchain signatures that ensure integrity and prevent unauthorized alterations. Generative AI and Natural Language Processing (NLP) are integrated to offer chatbot-based health assistance, symptom explanation, and personalized guidance. The application also includes educational video content and government healthcare scheme information through WebView modules ensuring that users have access to trustworthy support resources. The project combines AI cognition, immutable blockchain security, and real-time communication to deliver a reliable digital healthcare framework that enhances trust, accessibility, and patient doctor collaboration.

Keywords: Blockchain, SHA-256, Android, Generative AI, NLP, Firebase Realtime Database, Digital Health Records, Cloud Storage.

Abstract: The rapid growth of internet technologies and the recent global shift toward online education have significantly increased the demand for efficient and scalable e-learning platforms. This research paper presents the design and implementation of an E-Learning Ecosystem that integrates students, instructors, and administrators on a single web-based platform. The proposed system aims to provide flexible, user-friendly, and cost-effective digital learning through recorded courses, live sessions, online assessments, secure payments, and discussion forums. Developed using Angular for the frontend and Spring Boot with Hibernate for the backend, the system ensures modularity, scalability, and secure data management. The results demonstrate that the proposed ecosystem effectively enhances accessibility, reduces manual effort, and improves the overall learning experience.

Keywords: E-learning, Learning Management System, Web Application, Angular, Spring Boot, Online Education

Abstract: This study presents a deep learning–based web mining framework to detect fake reviews and improve e-commerce recommendation quality by integrating textual, behavioral, user–item metadata, and temporal signals. Reviews are modeled as tuples u(i,t,x,s,y) and transformed into structured feature vectors ϕ(r) that concatenate behavior/context features, user and item profiles, rating signals (including deviation from item mean), and time-based features extracted from crawled e-commerce pages and logs. A multimodal fake-review detector combines a neural text encoder (e.g., Transformer) with engineered web-mined features to estimate p(fake∣r) and derive a credibility score c_r=1-p(fake) . This credibility is then used to down-weight suspicious reviews during review aggregation and recommendation learning, enabling a credibility-aware recommender that is more robust to spam and coordinated manipulation. The framework supports joint multi-task optimization of detection and recommendation objectives and evaluates performance using standard detection metrics (Precision/Recall/F1, ROC-AUC, PR-AUC) and ranking metrics (HR@K, NDCG@K).

Keywords: Web mining, fake review detection, deep learning; transformer encoder, credibility scoring, multi-modal fusion, e-commerce recommender systems, joint learning

Abstract: This literature review analyzes the limitations of deep learning-based weather forecasting models. Models such as LSTM, BiLSTM, GAN-LSTM and FBVS-LSTM currently used are generally trained on single-location and sparse data, which limits their scalability and transferability to other climate domains. Most studies lack transfer learning, multi-location validation and real-time implementation. Long-term forecasting, feature importance analysis and interpretability have also been ignored. Rainfall forecasting still needs improvement, particularly during times of high variability. Furthermore, the majority of models do not make use of contemporary architectures like transformers and are instead based on outdated DL frameworks. . The study also shows that sentiment and situation-aware forecasting cannot be done with single-domain models. Future studies should focus on interpretable, adaptive, and real-time forecasting systems, multi-location data, and contemporary DL methodologies.

Keywords: LSTM, CNN, GA, RNN, NWP, ANN.

Abstract: Recent advances in the field of artificial intelligence (AI) have created new ways for machines to process information, moving from tasks that focus on analyzing and distinguishing data to more complex and creative activities using generative AI. By using deep generative models, generative AI can create new and realistic content like text, images, or code in different areas, based on simple user instructions. This paper offers a complete introduction to the basics of generative AI, including its key ideas and future possibilities. This paper explain important terms and methods, describe the main features of generative AI, and discuss its opportunities and challenges. It stresses the importance for researchers and practitioners to understand the unique aspects of generative AI so they can use its strengths, manage its risks, and help build a better understanding of it.

Keywords: Generative AI, Artificial intelligence, Deep learning, Deep generative models, Large language model.

Abstract: Head and neck cancer ranks among the most common and deadly forms of cancer globally. Detecting the disease at an early stage is essential for increasing survival rates and improving treatment success. Conventional diagnostic approaches typically depend on manual examinations and invasive testing procedures, which can contribute to delayed identification and higher mortality rates. This study introduces a machine learning–driven predictive framework designed for the early detection of head and neck cancer using clinical information. The model incorporates demographic characteristics, lifestyle habits, prior medical conditions, and reported symptoms to build an automated and effective prediction system. To enhance performance and minimize irrelevant data, the dataset undergoes preprocessing steps such as data cleansing, normalization, and feature selection. Several machine learning techniques—including Logistic Regression, Support Vector Machine, Decision Tree, Random Forest, and K-Nearest Neighbours—are applied and compared to determine the most accurate classification model. Evaluation is carried out using standard performance indicators, including accuracy, precision, recall, and F1-score. The findings reveal that ensemble learning methods outperform traditional classification algorithms in predictive capability, demonstrating their effectiveness for medical diagnostic tasks. The developed system is intended to support healthcare practitioners in making timely clinical decisions, lowering diagnostic inaccuracies, and improving overall efficiency. Ultimately, this research highlights the promise of machine learning in creating dependable, non-invasive, and affordable solutions for head and neck cancer prediction, contributing to enhanced patient outcomes and smarter healthcare systems.

Abstract: Agricultural commodity prices in India's Vidarbha region (Maharashtra)—a key producer of cotton, soybean, oranges, and pulses—exhibit extreme volatility due to erratic monsoons, seasonal supply-demand imbalances, transport logistics, limited storage, and government policies like minimum support prices (MSP). This unpredictability causes financial distress for over 1.5 million farming households, traders, and consumers, who lack reliable forecasting tools beyond rudimentary historical averages or linear statistical models. Such traditional approaches fail to model the non-linear, multifaceted patterns in price time series, especially during events like droughts or market surges. This research addresses the gap by developing and evaluating machine learning (ML) models for accurate price forecasting using historical data (2015–2024) from APMC mandis in Nagpur, Akola, and Yavatmal. After rigorous preprocessing (outlier removal, normalization, and feature engineering with lags, weather, and arrivals), we trained regression models—linear regression, support vector regression (SVR), random forest, XGBoost, and LSTM—on chronologically split datasets. XGBoost emerged superior (test MAPE: 4.8%, R²: 0.92 for soybean), outperforming ARIMA by 60% and capturing Vidarbha-specific volatilities. LSTM excelled in long-term dependencies. These results validate ML's potential for nonlinear time series analytics, providing farmers actionable predictions for crop planning, harvest timing, storage, and sales. Deployable via mobile apps with APMC APIs, this framework enhances decision-making, stabilizes incomes, and supports policy in climate-vulnerable regions. Future enhancements include real-time satellite integration.

Keywords: Vidarbha Agriculture, Price Forecasting, Machine Learning, XGBoost, Time Series, Commodity Volatility.

Abstract: Intelligent Transportation Systems (ITS) integrate advanced sensing, communication, and computational technologies to enhance traffic efficiency, safety, and sustainability. With the rapid growth of urban traffic, traditional traffic management approaches are no longer sufficient to handle dynamic and complex transportation scenarios. This paper presents an intelligent transportation framework that utilizes Machine Learning and optimization algorithms for real-time traffic analysis and decision-making. K-means clustering is employed to identify traffic density patterns, while Decision Tree and Random Forest algorithms are used for traffic congestion and accident prediction. Shortest Path algorithms such as Dijkstra and A* are applied for dynamic route optimization based on real-time traffic conditions. Simulation results demonstrate that the proposed approach reduces average travel time, improves traffic flow, and enhances road safety. The study highlights the effectiveness of algorithm-driven ITS solutions in supporting smart city transportation and sustainable mobility.

Keywords: Intelligent Transportation Systems, Machine Learning Algorithms, Traffic Prediction, Route Optimization, Smart Mobility.

Abstract: The rapid proliferation of AI-generated content, including synthetic text, images, and multimedia, has created significant challenges in digital trust, academic integrity, and online misinformation. Traditional detection mechanisms struggle to differentiate between human-written and AI-generated content due to improvements in large language models (LLMs). This study proposes a hybrid deep learning framework for AI-generated text detection by integrating transformer-based contextual embeddings with linguistic and stylometric features. The proposed model combines RoBERTa embeddings with statistical linguistic markers and employs an ensemble classifier to improve robustness. Experimental evaluation on benchmark datasets demonstrates improved detection accuracy compared to baseline transformer-only approaches. The proposed method achieves 96.3% accuracy and shows strong generalization across unseen AI models. The results highlight the importance of hybrid modeling for reliable AI content authentication.

Keywords: AI-generated content, fake content detection, deep learning, stylometry, transformer models, misinformation detection

Abstract: Gas leakage poses a significant threat in residential and industrial environments, often leading to hazardous situations like fires and explosions. This paper presents a Smart IoT-based Gas Leakage Detection and Auto Shut-off System that enhances safety by integrating real-time monitoring with an automated response mechanism. The system detects gas leaks using an MQ-6 sensor and promptly shuts off the gas supply through a servo motor-based valve control. Additionally, an alert notification is sent to users via an LCD display and buzzer to ensure immediate awareness. This paper discusses the half implemented version of the system, focusing on gas detection and auto shut-off functionalities. The proposed system improves upon existing gas detection techniques by offering a cost-effective, highly responsive, and automated approach to preventing gas-related accidents.

Abstract: Emotion recognition has become a significant research area in affective computing and human–computer interaction, as understanding human emotions plays a vital role in developing intelligent and responsive systems. Traditional unimodal emotion recognition systems rely on a single source of information such as speech, facial expressions, or text, which often leads to limited performance due to the absence of complementary contextual cues. To overcome these limitations, multimodal emotion recognition integrates multiple modalities—typically audio, visual, and textual data—to capture a more comprehensive representation of human affective states. This paper presents an attention-based deep neural network framework for multimodal emotion recognition. The proposed approach leverages deep feature extraction techniques using Convolutional Neural Networks (CNNs) for visual data, Recurrent Neural Networks (RNNs)/Long Short-Term Memory (LSTM) networks for audio sequences, and contextual embedding models for textual information. An attention mechanism is incorporated to dynamically assign weights to the most informative features across modalities, enabling the model to focus on emotionally salient cues while reducing irrelevant noise. The fusion of multimodal features is performed through a hybrid attention-based integration layer, enhancing the robustness and generalization capability of the system. The proposed model aims to improve classification accuracy across standard emotion categories such as happiness, sadness, anger, fear, and neutrality. Experimental evaluation on benchmark multimodal emotion datasets demonstrates that the attention-based fusion strategy significantly outperforms traditional unimodal and early-fusion approaches. The results highlight the effectiveness of attention mechanisms in capturing cross-modal dependencies and improving emotion prediction performance. This study contributes to the advancement of intelligent emotion-aware systems that can be applied in virtual assistants, mental health monitoring, smart education platforms, and interactive AI systems.

Keywords: Multimodal Emotion Recognition, Attention Mechanism, Deep Neural Networks, Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), Multimodal Fusion, Affective Computing, Speech Emotion Recognition, Facial Expression Analysis, Transformer Networks, Human–Computer Interaction, Cross-Modal Learning.

Abstract: Trip planning often involves using multiple platforms for destination research, map navigation, and manual itinerary organization, leading to fragmented and inefficient planning. Existing tools typically offer either basic AI-generated suggestions or isolated map-based exploration, lacking real-world feasibility and transparency. This paper presents Wanderly AI, an enhanced AI-powered personal travel assistant that integrates map-aware large language model (LLM) reasoning with full-stack web technologies. The system generates practical itineraries by incorporating geographic constraints such as distance, travel time, coordinates, and map context. To ensure consistency and reliability, reproducibility mechanisms using seeded prompts and version-controlled templates are implemented. Additionally, explainable AI techniques are used to justify itinerary decisions, including activity selection, sequencing, and timing based on user preferences and feasibility. The proposed system delivers realistic, personalized, and transparent travel planning, addressing key limitations of conventional trip-planning applications.

Keywords: Travel planning, Map-aware AI, Itinerary generation, Explainable AI, Reproducible AI, Geographic constraints

Abstract: This study examines the role of digital health technologies in the field of cancer care management, highlighting their potential benefits and associated challenges. A focused review of existing cancer care practices reveals fragmented approaches to patient monitoring, treatment adherence, and health awareness, emphasizing the need for integrated digital solutions. The findings suggest that mobile health systems can improve patient engagement, enable continuous monitoring, and support informed medical decision-making, while also raising concerns related to data security, accessibility, and user adaptability. The impact of digital cancer care systems is influenced by contextual factors such as healthcare infrastructure, patient literacy, and clinical involvement. Based on these observations, this study proposes a supportive digital approach that enhances traditional cancer care through structured tracking and secure communication rather than replacing clinical processes. By addressing current gaps in cancer care delivery, this work contributes to a clearer understanding of technology-assisted healthcare and emphasizes the role of digital platforms in improving long-term cancer management [1]

Keywords: Cancer care management, Mobile health application, Patient monitoring, Treatment adherence, Digital healthcare, Telemedicine, Symptom tracking, Remote healthcare support, Health data management, Patient engagement

Abstract: Solid waste management remains a critical challenge in African cities, where rapid urbanization, poor infrastructure, and limited resources often lead to inefficiencies and environmental hazards. In Enugu State, Nigeria, waste collection is managed by the Enugu State Waste Management Authority (ESWAMA) through a network of bin drop centers serviced by trucks from central depots. The current practice relies on static scheduling, which results in overflowing bins, excessive operational costs, and delays due to traffic congestion and long routing distances. This paper proposes a Model Predictive Control (MPC) framework for smart waste collection routing tailored to the operational context of Enugu State. The system models waste collection as a dynamic vehicle routing problem and leverages predictive control to forecast bin fill levels, adapt to real-time traffic conditions, and optimize routing decisions. Simulations conducted across urban, semi-urban, and rural scenarios demonstrate that the MPC-based approach reduces total distance traveled by up to 23%, decreases collection times by over 20%, and lowers overall operational costs and emissions compared with static scheduling. The results highlight the potential of MPC to transform municipal waste management into an adaptive, efficient, and sustainable system for African cities. Future work will integrate Internet of Things (IoT)-enabled smart bins and machine learning-based forecasting to further enhance prediction accuracy and scalability.

Abstract: The nutritional value of mulberry leaves plays a pivotal role in sericulture, directly impacting silkworm development and silk production. Traditional methods of assessing leaf health rely on manual inspection, which is often subjective, labor-intensive, and impractical for large-scale monitoring. This paper introduces a real-time deep learning system for detecting nutrient deficiencies in mulberry leaves, combining YOLOv8-based instance segmentation with LAB color space clustering. The model efficiently detects and classifies nitrogen, potassium, and magnesium deficiencies by analyzing subtle color variations in leaf tissue. Experimental results show that the proposed system achieves superior accuracy and precision compared to conventional techniques. Moreover, it delivers rapid and scalable performance suitable for field-level deployment. To enhance usability, the detection model is integrated into a user-friendly interface, empowering sericulture farmers to make informed, data-driven decisions for improving leaf quality. This automated solution aims to increase productivity, reduce losses, and support sustainable silk farming through optimized nutrient management.

Keywords: YOLOv8, Instance Segmentation, Mulberry Leaves, Nutrient Deficiency Detection, LAB Color Space, Deep Learning, Sericulture, Sustainable Agriculture.

Abstract: Skin diseases are widespread and require timely detection to prevent complications. However, access to dermatological care is limited in many regions, leading to delayed diagnosis and treatment. Derma-Scan is an AI-powered Android application designed to provide users with fast and reliable preliminary skin disease detection using smartphone-captured images. The system integrates a deep learning-based Convolutional Neural Network (CNN) model optimized with TensorFlow Lite to analyse skin images and predict potential diseases with high accuracy. The application offers an intuitive interface, real-time prediction, secure data handling, and a history feature for tracking user reports. Derma-Scan aims to support early screening, increase accessibility to dermatological insights, and assist users in making informed decisions regarding further medical consultation. This project demonstrates the potential of mobile AI solutions in healthcare, particularly in resource-constrained environments.

Keywords: Skin Disease Detection, Artificial Intelligence, CNN, TensorFlow Lite, Android Application, Medical Imaging, Dermatology, Deep Learning.

Abstract: In the domain of modern security and surveillance, the delay between threat detection and response is a critical vulnerability. This paper presents AimSense, a computer vision-based threat detection system currently under development. The system utilizes the You Only Look Once (YOLO) version 11 (YOLOv11) architecture to integrate object detection (weapon recognition) with pose estimation (human skeleton analysis), enabling accurate identification of active threats based on grasping interactions rather than mere object presence. An important component of the proposed system is the Human-in-the-Loop (HITL) interface, which ensures that all the engagement decisions are verified by a human operator prior to execution. This paper describes the prototype architecture, the sector-based localization algorithm, and the optimization techniques that enable real-time performance on standard hardware.

Keywords: Computer Vision, YOLOv11, Threat Detection, Human-in-the-loop, Pose estimation, Surveillance Systems.

Abstract: Elections are an essential part of democratic systems, but traditional voting methods still face challenges such as identity fraud, duplicate voting, and difficulties in voter verification. Manual authentication processes are time-consuming and may lead to errors, reducing transparency and efficiency during elections. With the advancement of biometric technologies and artificial intelligence, secure electronic voting systems can be developed to improve voter authentication and prevent unauthorized voting. This paper proposes VOTEX – One Person One Vote, a biometric-based voting system that combines fingerprint verification and face recognition to ensure secure and accurate identification. The system captures voter details and biometric data during registration, verifies identity through multi-level authentication, and prevents multiple voting by maintaining real-time vote status in the database. It also provides flexible authentication options for disabled voters to ensure accessibility and inclusiveness. By integrating technologies such as Flask, OpenCV, MTCNN, CNN-based face recognition, and MySQL, the proposed system aims to provide a reliable, secure, and user-friendly voting solution that enhances electoral transparency and trust.

Keywords: Biometric authentication, electronic voting system, face recognition, fingerprint verification, MTCNN, CNN, Flask framework, secure voting, voter authentication, one person one vote

Abstract: The Intelligent Multimodal Notes Generation System is designed to simplify the process of note-making by integrating multiple input modes such as text, audio, and visual content. It leverages Artificial Intelligence and Natural Language Processing (NLP) to automatically analyze lectures, documents, or multimedia inputs and convert them into well-structured, concise, and context-aware notes. The system identifies key concepts, summarizes content, and organizes it in a user-friendly format, enhancing learning efficiency and retention. It supports features like speech-to-text conversion, summarization, keyword extraction, and diagram or image interpretation, making it highly beneficial for students, educators, and professionals. By reducing manual effort and ensuring accuracy, this system addresses the challenges of traditional note-taking and provides personalized, intelligent, and accessible digital notes.

Keywords: Artificial Intelligence, Multimodal Input, NLP, Note Generation, Summarization, Speech-to-Text, Educational Technology, Knowledge Extraction, Automation, Smart Learning System

Abstract: Street lighting is an essential component of modern urban infrastructure, ensuring road safety, pedestrian security, and improved public visibility during nighttime conditions. However, traditional street lighting systems operate using fixed schedules or constant brightness levels, regardless of environmental changes or traffic density. This results in excessive power consumption, higher electricity costs, and inefficient resource utilization. In developing regions, where energy management is critical, the need for intelligent and automated lighting systems has become increasingly important.This paper presents the design and implementation of a cost-effective Internet of Things (IoT)-based smart street lighting system aimed at enhancing energy efficiency and automation. The proposed system integrates environmental sensors, including a Light Dependent Resistor (LDR) for ambient light detection and an infrared (IR) sensor for vehicle or motion detection. A temperature sensor is also incorporated for environmental monitoring. These sensors are interfaced with a WiFi-enabled microcontroller that collects real-time data and transmits it to a centralized server for processing and storage.The system operates dynamically by analyzing ambient light intensity to determine day and night conditions. During daylight hours, the street lights remain switched off to conserve energy. At nighttime, the system adjusts brightness levels based on detected vehicle movement. When traffic density is low, the lights operate at reduced intensity, whereas brightness increases when movement is detected to ensure road safety. All sensor readings and lighting status information are stored in a MySQL database for monitoring, data analysis, and performance evaluation.Experimental observations demonstrate that the proposed IoT-based lighting system significantly reduces unnecessary energy consumption while maintaining adequate illumination standards. The system is scalable, affordable, and suitable for smart city applications. By minimizing human intervention and enabling real-time monitoring, the proposed solution contributes to sustainable urban energy management and improved infrastructure efficiency.

Keywords: Internet of Things (IoT), Smart Street Lighting, Energy Efficiency, Motion Detection, Ambient Light Sensor, Wireless Communication, Smart City.

Abstract: Anomaly detection plays a critical role in modern cybersecurity systems due to the increasing scale, complexity, and temporal nature of network traffic. Traditional intrusion detection systems often generate isolated anomaly alerts without providing a higher-level interpretation of entity reliability. To address this limitation, this paper proposes TrustCast, a trust-aware deep learning framework that integrates temporal anomaly detection with dynamic trust com- putation. TrustCast employs data augmentation to address class imbalance, a GRU-based sequential autoencoder for time-series anomaly detection, and a trust computation module that converts anomaly evidence into dynamically evolving trust scores. Experimental results demonstrate that TrustCast outperforms baseline models in detection accuracy while providing interpretable trust trajectories suitable for proactive security decision-making.

Keywords: Anomaly Detection, Trust Computation, Deep Learning, Cybersecurity, Time-Series Analysis.

Abstract: Cardiovascular diseases (CVDs) are the leading cause of global mortality, as reported by the World Health Organization. Early identification of individuals at high risk is essential for effective prevention and clinical intervention. However, conventional prediction models often analyze cardiac parameters independently and fail to consider the strong correlations between metabolic disorders such as diabetes, obesity, dyslipidemia, and hypertension.This study proposes a novel deep learning-based predictive framework that integrates metabolic disorder correlations for early CVD detection. The model utilizes multi-parameter clinical data, including blood glucose, body mass index, cholesterol levels, and blood pressure, to capture complex nonlinear relationships among risk factors. A correlation-aware neural network architecture is developed to enhance predictive performance and robustness.Experimental results demonstrate improved accuracy and ROC-AUC compared to traditional machine learning approaches. The proposed framework supports early risk stratification and provides a scalable solution for preventive cardiovascular healthcare applications. Keywords Deep learning, cardiovascular diseases (CVD), metabolic disorders, early detection, predictive modeling, artificial neural networks (ANN), risk stratification, explainable artificial intelligence (XAI).

Abstract: Cloud storage solutions often face large amounts of redundant data, leading to inefficient storage resource usage and increased operational expenses. Data deduplication helps to overcome redundancy, but the deduplication of encrypted data raises issues related to security and privacy. This paper proposes an efficient and secure multi-cloud data deduplication solution by combining Multi-Key Homomorphic Proxy Re-Encryption (MKH-PRE) with Elliptic Curve Cryptography (ECC). The proposed design uses Content Defined Chunking (CDC) to divide files into variable-length chunks and ECC-based hashing to generate secure identifiers for deduplication. A Distributed Hash Table (DHT) approach is used to ensure fair data replication among storage nodes, and a Proof-of-Ownership (PoW) method is used to verify authentic users. Performance evaluation shows improved storage resource usage, reduced computational complexity, and secure cross-cloud deduplication suitable for modern cloud setups.

Keywords: Cloud Security, Data Deduplication, MKH-PRE, Content Defined Chunking, Proof of Ownership.

Abstract: This article proposed a spatial-temporal deep learning architecture for network-wide flight delay prediction that operates within the confines of federated learning to ensure data privacy is respected among various aviation stakeholders. Due to regulatory, commercial and privacy limitations, it is not possible to use traditional centralized delay prediction models because airline operators, airports and air traffic control authorities have access to sensitive operational, passenger and meteorological data which cannot be aggregated in a centralised form. In order to tackle this fundamental problem, we propose the Hybrid Federated Delay Learning Network (HFDL-Net), which employs a spatio-temporal graph neural network with gated recurrent units (GRU) at each client node, in conjunction with an hierarchical federated aggregation approach at the central server to learn together from distributed datasets without direct data sharing. Using our architecture, we model the air transportation network as nodes with edges (aeroportunities) and routes with convolution (temporal evolution) via graph modeling and recurrent layers to capture spatial delay propagation patterns. Real-world experiments on multi-airport flight datasets spanning three major hub networks demonstrate that HFDL-Net achieves mean absolute error (MAE) improvements of 12–15% over non-federated baseline models while maintaining prediction accuracy within 3% of fully decentralized training approaches. In addition, the use of a hierarchical aggregation reduces communication overhead by 40% when compared to traditional FedAvg implementations through adaptive client selection and gradient compression techniques. The suggested scheme effectively manages non-IID data distributions among multiple clients, exhibits resilience to client dropout scenarios, and adapts well to airport networks spanning more than 100 participants. Additionally, This evidence supports federated spatial-temporal modeling as a practical, scaleable and privacy-preserving approach for network-wide flight delay prediction in real-life aviation scenarios where data sovereignty and regulatory compliance are critical requirements.

Keywords: Flight Delay Prediction, Spatial-Temporal Modeling, Federated Learning, Graph Neural Networks.

Abstract: The rapid growth of large datasets in industries such as healthcare and finance has created a strong need for smarter tools that can both visualize data automatically and check whether algorithms are fair. In the past, researchers have worked on data visualization and fairness separately, but there is no single system that combines both in one platform. This paper reviews more than 25 recent research studies covering automated chart recommendations, fairness-aware machine learning, explainable AI, and bias detection. From this review, we identify several important gaps. Current systems do not combine visualization and fairness analysis in a single framework, they lack real-time bias monitoring, and they are often difficult for non-technical users to understand and use. To address these issues, we propose and develop a prototype called the AI-Powered Automated Data Visualization and Fairness Analysis Platform (AADVFAP). Our prototype shows that it is possible to build an integrated platform that handles both visualization and fairness analysis effectively. The proposed system is modular and scalable, and it is designed to support data scientists, domain experts, and policy decision-makers.

Keywords: Data Visualization, Algorithm Fairness, Bias Detection, Fair Machine Learning, Explainable AI, AI-Based Data Analysis, Real-Time Bias Checking, User-Friendly Analytics, Ethical Artificial Intelligence.

Abstract: Epileptic seizure detection is one of the critical challenges in biomedical signal processing, with inherent non- linearity and noise in EEG recordings. While traditional machine learning and deep learning approaches have achieved promising accuracy, generalizability, class imbalance, and interpretability remain concerning issues. The objectives of this work are to develop an ensemble-based approach by combining RF with BTC to improve the robustness and sensitivity of seizure detection. The pre-processing steps have been done using normalization and label encoding on raw EEG data obtained from the UCI Epileptic Seizure Recognition dataset. The hybrid model combines the merits of both RF, which reduces variance with the large randomness provided by the features, and BTC, with its reduced overfitting via bootstrap aggregation. These experimental results demonstrate that the proposed hybrid method performs superiorly when compared to individual ML models in several performance metrics: accuracy, precision, recall, and F1-score. The proposed work is a computationally efficient, interpretable, and reliable seizure detection framework for real-world and portable EEG monitoring systems.

Keywords: Epileptic seizure detection, EEG signals, hybrid machine learning, Random Forest, and Bagged Tree Classifier.

Abstract: Parking congestion in urban and semi-urban areas has become a critical issue due to the rapid increase in vehicle population and inefficient parking management. This paper presents the design and implementation of an IoT-enabled smart parking system for both indoor and outdoor organized parking areas. The proposed system uses ultrasonic sensors installed at each parking slot to detect vehicle occupancy. An ESP32 microcontroller acts as a zone controller, collecting sensor data and transmitting it via Wi-Fi to a cloud-based backend. The system employs a Node.js server for data processing and MongoDB Atlas for cloud data storage. A web-based dashboard developed using React and Vite provides real-time visualization of parking slot availability using a grid-based layout. Experimental results show that the proposed system provides accurate real-time slot monitoring, reduces parking search time, and improves parking space utilization. The system is scalable, cost-effective, and suitable for deployment in smart parking environments such as malls, campuses, hospitals, and public parking facilities. Keywords— Smart Parking, IoT, Ultrasonic Sensor, ESP32, MongoDB Atlas, Node.js, Web Dashboard.

Abstract: The rapid growth of internet-based services, cloud computing, and interconnected network infrastructures has significantly increased the risk of cyberattacks and unauthorized access. Traditional intrusion detection systems (IDS), which rely primarily on signature-based or rule-based techniques, often fail to detect newly emerging or sophisticated attack patterns. These limitations highlight the need for intelligent and adaptive security mechanisms capable of analyzing large volumes of network traffic and identifying malicious activities in real time. To address this challenge, this study proposes an Artificial Neural Network (ANN)-based approach for intelligent network intrusion detection that enhances the accuracy and efficiency of cybersecurity monitoring systems. The proposed framework utilizes the learning and pattern recognition capabilities of artificial neural networks to analyze network traffic data and classify it into normal or malicious categories. The multilayer neural network architecture is designed to capture complex relationships within network features and detect anomalies that indicate potential cyber threats. The system performs data preprocessing and feature extraction to improve the quality of input data and reduce noise and redundancy. The ANN model is then trained using benchmark intrusion detection datasets containing various types of network attacks, including denial-of-service (DoS), probing attacks, remote-to-local (R2L), and user-to-root (U2R) intrusions. Experimental results demonstrate that the proposed ANN-based intrusion detection model provides improved detection accuracy, higher precision, and lower false alarm rates compared with conventional intrusion detection techniques. The adaptive learning capability of neural networks enables the system to identify previously unseen attack patterns and continuously improve its performance over time. Furthermore, the framework supports real-time monitoring and scalability, making it suitable for deployment in modern network environments such as enterprise networks, cloud computing platforms, and Internet of Things (IoT) systems. This research highlights the effectiveness of artificial neural networks in strengthening network security by providing an intelligent and automated mechanism for detecting and preventing cyber intrusions in next-generation network infrastructures.

Keywords: Artificial Neural Network (ANN); Network Intrusion Detection; Cybersecurity; Intrusion Detection System (IDS); Machine Learning; Network Security; Anomaly Detection; Cyber Attack Detection; Intelligent Security Systems; Deep Learning in Security

Abstract: In today’s digital environment, organizations require efficient and automated customer support systems to handle inquiries, bookings, and information requests without human dependency. Traditional receptionist services are limited by availability, response time, and operational cost. To overcome these limitations, this project proposes a Virtual Receptionist Chatbot System that provides automated, intelligent, and real-time interaction with users. The system integrates a web-based chatbot interface, Natural Language Processing (NLP) engine, and cloud-based database such as Firebase to deliver accurate responses to user queries. The chatbot processes text or voice inputs, retrieves relevant information from the knowledge base, and generates responses in the form of text, images, forms, or booking confirmations. Additionally, the system stores conversation data in the cloud for monitoring, analytics, and service improvement. This solution enhances customer engagement, reduces waiting time, and provides 24/7 automated assistance while supporting organizations in improving operational efficiency and decision-making.

Abstract: Doomscrolling is an emerging behavioural pattern characterized by prolonged, repetitive, and low-attention smartphone usage, which often leads to reduced productivity and cognitive fatigue, particularly among students. Existing digital well-being solutions largely rely on static screen-time limits and manual restrictions, offering limited adaptability to individual user behaviour. This paper presents a prototype AI-based system for detecting and mitigating doomscrolling through behavioural analysis of smartphone usage metadata. The proposed approach passively monitors user interaction patterns during an initial learning phase and constructs a personalized behavioural profile using lightweight machine learning techniques. Due to the lack of publicly available labeled datasets for doomscrolling behaviour, a proxy-labeled dataset is generated using behaviour-inspired heuristics to evaluate model feasibility. A lightweight neural network model is trained to predict the likelihood of a user entering a doomscrolling state based on features such as session duration, scrolling intensity, application switching behaviour, and time-based usage patterns. To preserve user privacy and usability, the system operates without content inspection and deploys the trained model using on-device inference. Conservative decision thresholds and non-intrusive interventions are employed to minimize false-positive alerts and support user awareness rather than enforce restrictive controls. Experimental evaluation demonstrates effective behavioural separability under controlled conditions, validating the feasibility of the proposed approach. The results indicate that predictive, behaviour-aware digital well-being systems can provide a promising foundation for addressing doomscrolling in real-world mobile environments.

Keywords: doomscrolling, screen addiction, mobile usage analysis, machine learning, digital well-being.

Abstract: Agriculture is the most important developing sector. But nowadays, it consumes large amounts of water and electricity for the production because of the manual irrigation and scheduling methods [1], [6]. With the increase in population, farmers need to manage more land to irrigate but due to water scarcity, and an increase in power cost that affects more cost for the irrigation [4], [10]. This paper proposes an IoT and Machine Learning-based optimization system by providing water at the right time to the fields with low consumption of the electricity bill [3], [8]. Our proposed solution uses a Random Forest Regression model and it integrates with the IoT for the development. Here, our model works by taking the data from the ESP8266 NodeMCU microcontroller, soil sensor and water level sensor [5], [7]. From these sensors data is taken continuously over a 30-day period and using this we can predict the exact need of water to the fields. All the results are stored in the FireBase while using the wifi. The implemented model is used for controlling the manual operations for providing water to the fields and it helps to automatically turn ON or OFF when it reaches the required threshold value. This model mainly helps in reducing the water wastage, electricity cost that helps in improving the production. The implemented system achieves the water reduction wastage by 35% and electricity costs by 42% when compared to traditional fixed-irrigation systems, this helps in high model accuracy of (R2) of 0.94 [9], [10].

Keywords: Smart Irrigation, Soil Moisture Sensor, Water Level Sensor, ESP8266 Nodemcu Controller, Resource Efficiency, Energy Scheduling.

Abstract: The rapid growth of digital documents has been increasing exponentially in various domains and the effective retrieval of documents has become a challenging task. Traditional keyword search could not understand the real meaning of words. It only looks for matching words, so many times the results are not correct. It also misses the context of the question. Because of this, the answers are not accurate. Now, new technologies like Generative Artificial Intelligence and Natural Language Processing make this work easier and better. The objective of our project is to develop an intelligent document querying system that enables efficient question-answering by integrating document retrieval methods with Large Language Models through a Retrieval Augmented Generation (RAG) framework. This system focuses on context-aware answer generation by including semantic search techniques rather than typical keyword-based retrieval. Our proposed system supports multiple document formats ingestion like PDFs, Text files or Docs. These documents are divided into multiple segments known as embeddings and are stored in a Vector Database which enables faster retrieval. When a user submits a query, relevant segments of the content are forwarded to the Language Model for accurate answer generation. The system is implemented using Python for backend development, vector indexing techniques for semantic retrieval, pretrained embedding models for representation learning, and Large Language Models for answer generation. The modular architecture ensures scalability and allows the system to be adapted to different document domains with minimal modification. The expected outcome of this project is a reliable question answering system, improving reliability, reducing ambiguity and minimizing hallucinated outputs usually associated with regular Language Models. The proposed system can be effectively deployed in educational institutions, research environments, legal documentation systems, and enterprise knowledge management platforms to support intelligent, data-driven decision making. Index Terms: Retrieval-Augmented Generation, Generative Artificial Intelligence, Natural Language Processing, Semantic Search, Vector Database, Document Question Answering, Large Language Models

Abstract: In today’s modern world, unhealthy eating habits and busy lifestyles have become major contributions to health issues like obesity, malnutrition and chronic diseases. Even though there are many standard nutritional guidelines many individuals fail to follow because of various reasons [1],[2]. The reasons may vary from person to person such as based on their age, previous health conditions, personal choices etc. So they find it difficult to follow and adapt generalized nutritional plans that are available. Many users are looking forward to having their own personalized nutritional plan rather than any other generalized nutritional plans. To overcome this challenge, this research proposes a customized nutritional plan for every individual by collecting data from user inputs like age, gender, height, weight, health conditions, personal choices, tastes etc[3].[5],[7]. Our project is aimed to use a rule-based system that gets predefined rules from the standard nutritional guidelines which are useful to make a free from danger and clinically accepted diet plans for each BMI category by filtering unsuitable, unhealthy foods and maintaining proper nutritional balance [4], [6]. The result of our project is a user-friendly and smart nutrition recommendation system that gives accurate, personalized, and preference-aware diet plans. By combining health data evaluation with NLP techniques, our project helps users follow healthier diets while still enjoying foods of their tastes and cultural preferences [5]. Index Terms: BMI Calculation, Personalized Diet Planning, Nutritional Recommendation, Rule-Based System, Large Language Model, Health NLP, User Food Preferences.

Abstract: The selection of foods is highly determined by the emotional and affective conditions of an individual, however prevailing food-recommendation systems are more concerned with the fixed preferences of the users, limitations of food intake or what has been previously consumed without taking into account the dynamic effect of mood [1][5]. In order to overcome this shortcoming, this paper introduces a Mood Adaptive Food Recommendation System which combines affective state analysis method with content-based filtering methods to produce personalized and context sensitive food recommendation [1]. The first step of the proposed framework is to recognize the current mood of the user with the help of affective characteristics based on self- reporting or emotion classification models [5]. The content-based filtering module uses food qualities like ingredients, nutritional value [5], the type of cuisine and health considerations to provide recommendations that are in line with the mood detected and individual food preferences [2]. In contrast to collaborative methods, the given method is not based upon a significant portion of user interaction data, which is effective in cold-start cases [6]. Temporal awareness is also integrated into the system to change the recommendation based on the shifting emotional patterns with time [3][4]. Index Terms: Mood Adaptive Recommendation, Food Recommendation System, Affective State Analysis, Content-Based Filtering, Emotion-Aware Personalization, Temporal Adaptation, Personalized nutrition, Affective computing, cold-start problem, Context-Aware Recommendation.

Abstract: Nowadays, companies are facing many problems because business conditions keep changing very fast. Customer needs can be varying from time to time which can result in change in competition, and market situations. Consequently, companies find it difficult to guess how much demand they will have in the future. If demand is not predicted correctly, it may result in extra stock or stock outs, which causes losses. So, demand forecasting has become an important part of supply chain planning. In this project, past sales data is used to predict future demand. The data is studied based on time to understand how demand changes in a certain period of time. Simply this is known as time- series forecasting. Two models(ARIMA, SARIMA) are used in this work. ARIMA is used when the data does not show seasonal changes, and SARIMA is used when demand repeats in a seasonal manner. The predictions obtained from these models help in planning inventory and purchasing activities. This allows companies to maintain enough stock without spending too much on storage. The method used in this project is easy to apply and is suitable for small and medium businesses. Overall, this project explains how previous sales data can be used in a practical way to support better planning and decision-making. Index Terms: Time-Series Forecasting, Demand Forecasting, Supply Chain Optimization, ARIMA Model, SARIMA Model.

Abstract: Imperfections in software lead to critical problems in software reliability and stability. Detection of errors at a early stage assures a decrease in software development costs and efforts. Conventional static software analysis tools function on predefined rules and processes. These are not always effective in unearthing any fundamental errors in code logic. As complexities in software evolve, better means of analysis are needed. Artificial Intelligence opens up new model of code analysis to the program- mer. This project is all about the automated code analysis of static code via advanced learning algorithms. This code is converted into structured forms that show how the code logic and syntax map. These structured forms aid the system in analyzing code relationships. Code defects are associated with patterns that a language model understands. The system is known to point to lines of code which are mostly liable to flaws. This is meant to help the programmer concentrate their efforts. The process can be used to test various projects. The project helps to ensure the production of quality software. Index Terms: Static Code Analysis, Defect Prediction, Large Language Models, Program Representation, Software Quality.

Abstract: In recent times touchless interaction has become useful in many areas like hospitals, public systems and smart environment, touching keyboards or screens is not always safe it can also create hygiene problems because of this there is a need for systems that work without touch this project is about developing a touchless system control using computer vision a normal camera is used to capture live video of the user hand, the system detects the hand and follows the movement of the fingers by finding the distance between fingers and different hand gestures are identified these gestures are used to perform simple system tasks such as moving the mouse cursor, clicking, scrolling, zooming and controlling volume no special devices are needed for this system which makes it easy and low cost. The system is simple to use and works in real time. It can be used in healthcare smart systems and other applications where touchless control is required. Index Terms: Touchless System Control, Computer Vision, Hand Tracking, Finger Distance Estimation, Gesture Recogni- tion, Human-Computer Interaction.

Abstract: Transforming an image from one form to another is nothing but image-style translation. The image transformation plays a key role in computer vision and digital media, especially in transforming real-world images to artistic styles such as cartoons. Generally these conversions can be done manually by artists; however, advancements in technology like image editing tools have made it easy. But, most of them are time-consuming and do not ensure that the originality of the image is preserved. Our project“A Generative Adversarial Network-Based Framework for Photorealistic -to-Cartoon Image Style Translation” focuses on generating the quality image translation, in addition to preserving the originality of the image. In this model we are using a CycleGAN, a deep learning framework which consists of a generator for generating the real to cartoon images and a discriminator for evaluating the generated output. The system mainly focuses on simplifying image textures, smoothening the surfaces and broadening the edges for cartoon-like appearance. The results of the system successfully generates the cartoon- style images that preserves the important features of the original image.This image transformation framework can be applied in areas such as digital art creation, animation preprocessing, social media filters, and creative design tools. Index Terms: Generative Adversarial Networks (GAN), Image Style Translation, Cartoon Image Generation, Photorealistic Images, Computer Vision, Deep Learning, Automated Image Transformation.

Abstract: Carbon emissions are one of the major contributors to global climate change and environmental degradation. Accurate estimation of carbon footprints is essential for promoting sustainable living and responsible industrial practices. Most existing carbon emission calculation systems rely on generalized regional averages, which often fail to provide accurate insights into emissions produced by individual households and industries. This paper proposes a Smart Segregated Emission Analytics Framework that calculates carbon emissions separately for residential and industrial sectors. The proposed system utilizes a Decision Tree machine learning algorithm to analyze energy consumption and predict emission levels effectively. By separating emission data sources and applying intelligent prediction techniques, the system provides more accurate emission estimates and helps identify major pollution contributors. The framework improves prediction accuracy, reduces computational complexity, and supports better environmental decision-making. The proposed system offers a cost-effective and scalable solution for monitoring and reducing carbon emissions in modern societies.

Keywords: Carbon Footprint, Carbon Emission Prediction, Decision Tree Algorithm, Machine Learning, Environmental Sustainability.

Abstract: Humanoid robots play an important role in modern education by providing an interactive and practical learning platform for students. Unlike conventional educational robots, humanoid robots are capable of mimicking human movements, which enhances student engagement and understanding of robotics concepts. This paper presents the design and control of a humanoid robot developed using the Arduino Mega 2560 Pro Mini for educational applications. The proposed humanoid robot consists of multiple servo motors to achieve articulated movement in the head, arms, and legs. The mechanical structure of the robot is designed using Fusion 360, allowing modular construction and easy modification. The Arduino Mega is selected as the main controller due to its ability to handle multiple servo motors simultaneously. The robot demonstrates stable motion control and synchronized joint movement, making it suitable for laboratory demonstrations and robotics education [1].

Keywords: Humanoid Robot, Arduino Mega, Educational Robotics, 19-DoF Kinematics, STEM Education

Abstract: Software complexity metrics provide quantitative mechanisms to evaluate essential quality attributes of software systems, including maintainability, testability, reusability, and overall design quality. Within Object-Oriented (OO) design, several metrics have been proposed to quantify structural and cognitive aspects of software complexity. Among these, Cognitive Weighted Inheritance Class Complexity (CWICC) and Interface-Based Cognitive Weighted Class Complexity (ICWCC) have been introduced to measure the cognitive burden associated with inheritance hierarchies and interface-oriented architectural constructs. This study presents a focused theoretical validation of both CWICC and ICWCC metrics using Abreu’s validation criteria, a well-recognized framework for assessing the soundness and appropriateness of object-oriented design metrics. The validation process systematically examines whether these metrics satisfy Abreu’s established properties for meaningful software measurement, including consistency, monotonicity, and proper representation of design characteristics. By evaluating CWICC and ICWCC against these criteria, the research aims to establish their theoretical robustness, strengthen their measurement credibility, and support their applicability in assessing cognitive complexity within object-oriented software systems.

Keywords: Complexity, Cognitive, Object-oriented metrics, Abreu’s validation criteria, Metric validation.

Abstract: Football is one of the most popular sports worldwide and involves high-intensity physical activity that exposes players to a considerable risk of injury. The present study aimed to examine the prevalence and time situation of injuries among elite football players. A descriptive retrospective research design was adopted for the study. A total of 1000 elite football players aged between 14 and 30 years were selected using purposive sampling from clubs, universities, and state teams affiliated with the All India Football Federation. Data were collected using a self-developed football injury questionnaire modified from Singh (2012). The collected data were analyzed using descriptive statistics and percentages through SPSS version 16. The results revealed that the highest percentage of injuries occurred during training/practice sessions (38.12%), followed by the second half of matches (30.88%), the first half of matches (25.33%), warm-up/conditioning (4.55%), and warm-down/cooling-down phases (1.33%). The findings highlight the importance of proper training management, injury prevention strategies, and structured warm-up and recovery protocols to reduce injury risks among football players.

Keywords: Football injuries, Elite football players, Injury prevalence, Training injuries, Match injuries, Sports injury prevention

Abstract: Plants play an important role in maintaining ecological balance and supporting human life. However, maintaining healthy plants requires proper care, timely monitoring, and early disease detection. Many plant owners lack the knowledge and tools needed to identify plant diseases and manage plant care activities effectively. This project proposes GreenCare, a smart web-based platform designed to assist plant enthusiasts in managing plant care and detecting plant diseases using artificial intelligence. The platform integrates modern web technologies and machine learning to create an intelligent plant management system. The system is developed using the MERN stack, which includes MongoDB, Express.js, React.js, and Node.js. GreenCare allows users to register, create profiles, share plant-related posts, and interact with other users through likes and comments. The system also includes a plant care reminder module that helps users schedule activities such as watering, fertilizing, and pruning plants. A key feature of the platform is the AI-based plant disease detection system, where users upload images of plant leaves. The system analyzes these images using a machine learning model developed with TensorFlow and Google Teachable Machine to identify plant diseases and provide predictions along with confidence scores. The GreenCare platform demonstrates how full-stack web development and machine learning technologies can be integrated to create an intelligent plant care management system that supports plant health monitoring, community interaction, and efficient plant care scheduling.

Keywords: Plant Disease Detection, MERN Stack, Machine Learning, Plant Care Management, Image Classification, Smart Agriculture.

Abstract: Leukemia, a severe form of blood cancer, requires immediate diagnosis and early intervention for higher survival rates. Traditional diagnostic methods involve the manual examination of blood smear slides under a microscope by expert pathologists, which is time-consuming, expensive, and prone to human error due to fatigue. In rural and resource-limited areas, the lack of digital scanners and specialized hematologists heavily delays the diagnosis. This paper proposes "BloodAI Pro," an automated, cost-effective "Lab-on-a-Phone" diagnostic system. The proposed system integrates a traditional microscope with a smartphone interface, utilizing a Hybrid Ensemble technique that combines Computer Vision and Deep Learning. A Convolutional Neural Network (CNN) trained on the C-NMC dataset forms the core prediction engine, while a custom mathematical Image Processing algorithm calculates Cell Density (Hypercellularity) to eliminate False Positives. Furthermore, the system is backed by a robust FastAPI server and automatically generates NABL-standard diagnostic PDF reports. The results demonstrate high precision, an extremely low False Negative rate, and the ability to classify the severity of the disease based on clinical cell-crowding logic.

Keywords: Leukemia Detection, Convolutional Neural Networks (CNN), Deep Learning, Image Processing, Hypercellularity, FastAPI, Computer Vision, Lab-on-a-Phone.

Abstract: The rapid growth of cybercrime and dark web activities has made threat monitoring an essential task for cybersecurity analysts. Many organizations struggle to identify potential cyber threats quickly due to the large volume of unstructured textual data generated across various online sources. This project introduces ThreatSpeak, a machine learning-based system designed to analyze and classify cyber threat information from textual data. The system allows users to upload threat-related text files, which are automatically preprocessed and analyzed using natural language processing techniques. A trained machine learning model categorizes the content into relevant cyber threat types such as phishing, malware, or data breaches. In addition to classification, the system extracts threat indicators, identifies potential targeted assets, and generates a concise analytical summary to assist cybersecurity professionals in understanding the threat context. The application is implemented using Python, Streamlit, and Scikit-learn, providing a simple web interface for threat analysis. ThreatSpeak demonstrates how machine learning can support cybersecurity intelligence by improving the speed and accuracy of threat identification from textual sources.

Keywords: Cybersecurity, Threat Intelligence, Machine Learning, Natural Language Processing, Dark Web Monitoring

Abstract: The problem becomes clear when you enter a government hospital that experiences heavy traffic on Mondays. The issue exists since long ago but continues to create problems. The majority of hospitals use either paper registers or basic digital token dispensers which operate on a first-come, first-served basis. The two methods fail to handle urgent medical needs which leads to actual patient care delays and creates unnecessary waiting periods for essential medical treatment. HopeStream was developed to fill this requirement. The software functions as a complete web-based hospital queue management system which unifies patient registration with triage-based priority classification and intelligent scheduling into a single system. Rather than assigning tokens in the order patients arrive, HopeStream asks a few targeted questions — body temperature, reported symptoms, and a short checklist of emergency indicators — and uses that information to assign each patient a priority level of High, Medium, or Low. The system performs this task without requiring the receptionist to decide about patient cases. The scheduling logic uses a weighted round-robin algorithm which maintains a 2:1 rule that High priority patients receive treatment after two High priority patients have been treated. The simple rule enables urgent cases to advance while still allowing doctors to treat patients who scheduled routine check-ups. The system operates department-wise to generate tokens while displaying a waiting-area display that updates every five seconds which includes a doctor panel and automated visit history functions that developers built using Python Flask and SQLite and Bootstrap 5.

Keywords: Hospital Queue Management, Priority-Based Scheduling, Weighted Round-Robin Algorithm, Triage Classification, Python Flask, SQLite, Real-Time Display, Role-Based Access Control

Abstract: Cloud computing has become an essential part of modern software systems, but managing the cost of cloud services is often difficult for organizations. Companies typically receive their cloud billing information as large CSV reports that contain detailed usage and cost data. While these reports contain useful information, understanding them manually can be confusing and time-consuming. It becomes especially difficult to identify sudden cost increases or understand which services are responsible for higher spending. CloudWise is an intelligent cloud expense analysis system designed to simplify the process of understanding cloud billing data. Instead of manually examining spreadsheets, users can upload their billing dataset directly into the system. The application automatically processes the data and converts it into meaningful insights such as monthly cost trends, service-wise spending distribution, and abnormal cost spikes. The main feature of CloudWise is its ability to detect unusual cloud spending patterns using statistical anomaly detection. When the system identifies a sudden increase in cost for a specific month, it highlights that period and analyzes which cloud service contributed most to the spike. The system also predicts future cloud expenses using a regression-based forecasting model, allowing organizations to plan budgets more effectively. Developed using Python Flask, Pandas, NumPy, and SQLite, the system presents its analysis through an interactive dashboard with charts, cost breakdowns, and automated insights. By transforming complex billing reports into simple visual information, CloudWise helps organizations monitor cloud spending more efficiently and respond quickly to unusual cost behavior.

Keywords: Cloud Cost Optimization, Anomaly Detection, Z-Score, Linear Regression Forecasting, Python Flask, SQLite, Interactive Dashboard, Statistical Analysis

Abstract: Signature-based defences cannot detect novel cyber threats in real time. We present THREAT-DETECT, a containerised platform that unifies a three-model stacking ensemble (Random Forest, Bidirectional LSTM, and 1D-CNN) with a DAG-based automated incident response engine. A 29-dimensional feature vector spanning lexical, WHOIS, and network signals feeds all models; SHAP attributions provide per-prediction explainability; and an uncertainty-based active learning controller continuously improves model quality. The Playbook Engine translates scored threat events into DNS sinkholing, firewall rule injection, TheHive case creation, MISP IOC export, and Slack alerting via auditable, rollback-capable DAG playbooks. Evaluation on 487K labelled examples yields ensemble AUC-ROC 0.9970 and median end-to-end response latency of 711 ms.

Keywords: threat detection · ensemble learning · deep learning · active learning · incident response · SHAP explainability · DAG playbook · DNS sinkholing · MISP · TheHive

Abstract: This paper presents an empirical evaluation of serverless computing architecture through the development and deployment of a REST API-based todo management system using AWS cloud services. The system leverages AWS Lambda for serverless compute, Amazon API Gateway for HTTP endpoint management, and Amazon DynamoDB for NoSQL data persistence. Through systematic performance testing and cost analysis conducted over a 30-day operational period, we demonstrate the practical advantages and limitations of serverless architecture for web applications. Performance measurements show average response latencies of 65-78 milliseconds for warm Lambda executions across CRUD operations, with success rates exceeding 99.7%. Cost analysis reveals monthly operational expenses of approximately $1.33 for 50,000 requests, representing a 97% reduction compared to equivalent EC2-based infrastructure. The study addresses challenges including cold start latency mitigation, CORS configuration, and DynamoDB schema design for NoSQL environments. Implementation artifacts include five Lambda functions totaling 850 lines of Python code and a responsive web frontend built with vanilla JavaScript. This work contributes practical insights for developers adopting serverless technologies and validates theoretical serverless computing advantages through measurable real-world deployment metrics.

Keywords: Serverless Computing, AWS Lambda, REST API, Amazon DynamoDB, Performance Evaluation, Cost Analysis, Function-as-a-Service, Cloud-Native Architecture.