VOLUME 14, ISSUE 9, SEPTEMBER 2025

Abstract: The exponential growth of digital transformation in higher education has positioned cloud computing as a critical enabler of academic and research excellence worldwide. Cloud computing has transformed university resource management through Infrastructure as a Service (IaaS), providing scalable and flexible solutions. However, optimizing resource allocation remains challenging due to dynamic workloads and fluctuating user demands, often resulting in underutilization, overprovisioning, and increased operational costs. Traditional allocation strategies inadequately address evolving academic requirements, necessitating data-driven approaches. This paper explores machine learning techniques for optimizing resource allocation in university cloud IaaS environments. The objectives were to: analyze Random Forest classification for predicting resource demand and examine K-means clustering for identifying usage patterns and anomalies in resource utilization. A mixed-methods research design was employed, collecting data from four Kenyan public universities: Moi University, Masinde Muliro University of Science and Technology, Turkana University, and Alupe University. Stratified sampling represented institutions of varying sizes, while purposive sampling selected ICT administrators and directors. Data sources included interviews, system logs, historical usage reports, and open IaaS datasets, analyzed through machine learning and thematic analysis. Key findings demonstrate significant optimization potential. The Random Forest model achieved 87.6% accuracy in demand prediction, effectively identifying peak periods and anomalies. K-means clustering revealed four distinct usage patterns (low, medium, high, and variable), enabling strategic resource planning. The combined application of both techniques enhanced resource allocation efficiency by 17%, reduced system response time by 33%, improved availability to 98.2%, and decreased operational costs by 20.7%. The study concludes that machine learning approaches significantly optimize university cloud IaaS resource management. The complementary nature of supervised and unsupervised learning techniques provides comprehensive insights for effective resource allocation, with practical implications for cost reduction and performance improvement in higher education institutions.

Keywords: Cloud resource allocation, Infrastructure as a Service (IaaS), Random Forest classification, K-means clustering and Machine learning optimization.

Abstract: Mathematics anxiety is a pervasive global barrier to student achievement and engagement, particularly in primary education. This study describe the potential of robotic applications, implemented within blended learning environments, to mitigate mathematics anxiety and improve learning outcomes. A sequential mixed-methods design was employed, comprising a quantitative survey (N=150) and semi-structured interviews (N=25) with pre-service teachers, in-service teachers, and faculty members in te partner countries (Bulgaria Latvia Lithuania, Poland and Türkiye). Survey results demonstrated strong positive perceptions: 88.4% agreed that robotics enhances mathematical success, and over 90% believed it makes mathematics more enjoyable and motivating for anxious students. However, a significant confidence gap was identified; while 80% expressed a desire to use robotics, only 38.4% felt confident in their ability to program robots. Interview findings corroborated these results, emphasizing the limitations of traditional abstract teaching and the potential of robotics to provide tangible, engaging experiences in teaching counting, measurement, and data processing. In response, the project team developed ten modular curriculum units designed to integrate robotics into primary mathematics education within a blended learning framework. The findings indicate that robotics can significantly reduce mathematics anxiety, provided educators receive adequate training, resources, and institutional support to bridge the efficacy gap.

Keywords: Mathematics Anxiety, Modular Curriculum, Robotic Application, Primary School Education.

Abstract: Analog-to-digital conversion systems face critical challenges, including noise interference, signal degradation, and limited adaptability in varying environmental conditions. This research introduces a machine learning-integrated conversion and classification system that transforms how we process audio signals in the digital era. Our intelligent conversion model employs a three-level quantization approach (-1, 0, 1) with user-defined thresholds, seamlessly integrated with logistic regression classification for enhanced pattern recognition. The system dynamically adapts between operational modes based on signal amplitude characteristics, achieving superior performance, 99.9% classification accuracy - demonstrating exceptional signal interpretation capability, CD-quality audio processing at 44.1 kHz sampling rate with minimal distortion, strong noise immunity with SNR of 31.6 dB and THD+N of -31.6 dB, and Real-time adaptive processing through intelligent threshold-based categorization. The study adopted Agile Methodology, implemented using MATLAB and validated through a comprehensive confusion matrix analysis. This system represents a standard shift from traditional signal processing to intelligent, self-adapting conversion technology. This study bridges the gap between standard signal processing and modern machine learning, providing a scalable solution for next-generation digital communication systems that require high fidelity and intelligent adaptability, and the modular architecture allows each processing phase to be individually tested and optimized, making it appropriate for telecommunications, industrial automation, and consumer electronics applications.

Keywords: Analog Signal, Machine Learning, Digital Signal, Signal Processing, Logistic Regression, Audio Classification, Adaptive Conversion

Abstract: Fedzora is a federated learning framework designed to enable collaborative training of AI models for cybersecurity applications while preserving data privacy. The framework integrates secure aggregation, differential privacy, and model validation to allow organizations to train threat-detection models without exposing raw sensitive data. This paper briefly presents Fedzora’s architecture, methodology, and deployment considerations.

Keywords: Cybersecurity, AI, ML, Fedzora Project, Vulnerability Assessment, Ethical Hacking, Quantum-Resistant Cryptography.

Abstract: The proliferation of social media platforms has raised significant concerns regarding its impact on the mental health of students. Since the mid-2010s, research has consistently indicated a correlation between high daily screen time and an increase in adverse mental health outcomes, such as anxiety, depression, and psychological distress in adolescents. These challenges often stem from mechanisms like passive content consumption and upward social comparison, which can trigger envy and depressive symptoms. The consequences are significant, negatively affecting academic performance, sleep quality, and overall well-being, sometimes escalating to severe psychological distress, including thoughts of self-harm. In response, the field has increasingly adopted machine learning to analyse large-scale digital data for early risk detection. Addressing this need, our project proposes a deep learning framework to proactively identify students at risk. Following a multimodal approach that fuses self-reported and behavioural data, a neural network model was developed. It was trained on a comprehensive dataset comprising thousands of anonymized entries from student surveys and their social media activity metrics to classify mental health status. In performance evaluations, the proposed model achieved a classification accuracy exceeding 85%, a result consistent with state-of-the-art benchmarks for similar tasks that report accuracies and precision metrics in the 85-90% range. The findings validate the efficacy of using artificial intelligence as a scalable, non-invasive screening tool within educational institutions. This approach supports the implementation of ethically-grounded early warning systems that can connect at-risk students with crucial support services. Ultimately, this work demonstrates the potential of technology to mitigate the negative psychological effects of social media and foster a healthier, more supportive environment for students.

Keywords: Social media, Mental health, Students, Deep learning, Machine learning, Neural networks, Data analysis, Prediction model, Stress detection, Online impact

Abstract: Brain tumours are often considered one of the most aggressive types of cancer. Historically, they were identified using conventional deep learning methods via MRI. Currently, studies are transitioning to advanced models that can analyse MRI scans to identify and classify tumours. Tumours are formed by abnormal cell growth in brain tissue and can be benign or malignant. Since treatment effectiveness and survival rates can be improved with early identification, this paper focuses on supervised learning approaches, such as Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN), to provide better and faster early detection.

Keywords: Brain tumour, Convolutional Neural Network, Recurrent Neural Network, Deep learning, Magnetic Resource Imaging (MRI), Artificial Intelligence, Medical Imaging, Early Detection, Tumour Classification.

Abstract: Counterfeit money is a huge threat for the economy and society. Traditional ways of detecting fake notes works well in controlled environments, but they are not fast enough, cannot handle a lot of cases and are not always easy to use in real life. The rise of Artificial Intelligence (AI) offers powerful tools for automatic and better fraud detecting solutions. It helps in closing gap between simple checks and real-time verification. This study compares two AI methods for finding counterfeit currency. One method uses statistical features from banknotes and processes them with Logistic Regression and K-Nearest Neighbours (KNN). The other method uses images and OpenCV to check visual security features like watermarks and security strips. The results show that the feature-based method is better at accuracy and speed for structured data, while the image-based approach works well for real world situations like mobile-verification. The study also applies these methods to prevent fraud in electronic payment systems like UPI and mobile banking, showing how AI can protect both physical and digital transactions.

Keywords: Anti-Counterfeiting Currency Detection, Financial Fraud Prevention, Machine Learning, Image Processing, Logistic Regression, K-Nearest Neighbours, OpenCV, Digital Payment Security, UPI Fraud Detection, Artificial Intelligence in Finance.

Abstract: Cloud computing has become a game-changing technology in healthcare. It allows for scalable storage, real-time information sharing, and innovation based on data. This study looks at how cloud solutions affect healthcare data systems, focusing on the situation in India. It uses a synthetic electronic health record (EHR) dataset of more than 900 entries, along with quantitative analysis and visual insights. The paper highlights key benefits such as efficiency and innovation while also discussing challenges like privacy and regulatory issues. It offers practical suggestions to help healthcare providers set up secure and effective cloud-based systems to enhance patient care.

Keywords: Cloud computing, healthcare data management, electronic health records, healthcare innovation, data security, India

Abstract: The Indian judiciary is presently facing an unprecedented and worrisome backlog of over 50 million cases that greatly compromises the fundamental constitutional directive of delivering timely justice. While there do exist prevailing digital efforts focused on enhancing judicial system efficiency that stand out, like the pioneering e-Courts Project and National Judicial Data Grid (NJDG), these initiatives to date remain restricted in their impact. To reduce these shortcomings, this document proposes a holistic plan for a single, three-tier AI–Cloud framework that has specifically been developed to efficiently tackle these systemic inefficiencies that hinder the present judicial system. The pivot feature of this work goes in conceptualizing the New India Model (NIM). that deals with laying down outcomes and optimizing procedural efficiency. A simulation over a period of six months, with synthetic and sanitized datasets extracted from sample inputs of datasets from the NJDG, showed a stunning 30% decrease in time taken to clear cases from the backlog with automated implementation of triage procedures. In addition to that, it was found that our Community Operational Performance Model (COPM) obtained a remarkable F1-score of 0.88, that acts to confirm system predictive reliability. During evaluation for a period, our system enjoyed a remarkable availability rate of over 99.9% when run on a hybrid infrastructure of clouds, besides yielding outputs that are explainable with supporting SHAP-based rationales, thereby greatly contributing to end user faith in system trustworthiness overall. Through this proposed study use to utilize cloud-native scalability with explainable AI. Like other previous works focused only on digitization, this proposed framework signifies how AI and Cloud together can virtualize case-related data, Reduce- pendency, and enhance citizen trust in the judiciary mechanism.

Keywords: Artificial Intelligence, Machine Learning, Large Language Models, Random Forest, Convolutional Neural Network (CNN), Intelligent Triage Assistant (ITA), Graph neural networks (GNNs).

Abstract: The increasing urban population and industrialization have led to an exponential rise in waste generation, posing significant environmental and health challenges. Traditional waste management systems, relying heavily on manual labor, are inefficient and error-prone. This paper proposes a Smart Waste Segregation System that integrates Internet of Things (IoT) and advanced sensor technologies to address these challenges. The system automatically categorizes waste into dry, wet, and metallic categories using ultrasonic, moisture, and inductive proximity sensors and ensures efficient collection and recycling. By leveraging real-time monitoring and automation, the system minimizes human intervention, reduces resource wastage, and prevents environmental degradation. This research consolidates findings from previous studies, highlights innovative design improvements, and presents a modified, cost-effective, and scalable system aimed at revolutionizing urban waste management. Additionally, the system’s integration with IoT platforms enables better decision-making and promotes sustainable waste management practices.

Keywords: Smart Waste Management, IoT, Waste Segregation, Urban Sustainability, Sensor Technology, Automation, Environmental Protection.

Abstract: Kanglish is one of the common used mixed language, usually used in social media to convey messages, written using english letters that sounds in Kannada, similar to Hinglish. This research is done to analyze the sentiments of different words and sentences based on the input kanglish sentences using mBERT model which is an deep learning technique. Many sentences were collected from various media platforms to prepare a dataset labeled with different emotions. The data was divided to train, test and validation set to train and test the model. Navarasa- the nine different emotions that are potraided in classical dance with different expression can also be expressed in words. Total of 12 different emotions are being labeled and the sentiment prediction model can predict the emotion of statement. AdamW optimization, cross entropy loss and earling stopping were used to prevent overfitting. Evaluation was carried out to test the performance and its accuracy. A performance with 0.92 score was achieved with the confusion matrix that highlighted the model’s capacity to differentiate among different emotions. Gradio is used as a user interface. The results shows the potential of the transformer based model architectures for improving the sentiment analysis for the languages that are not well resourced.

Keywords: Kanglish, mBERT, AdamW optimization, cross entropy loss, early stopping, Navarasa sentiment prediction.

Abstract: Autonomous cars (AVs) are being developed at a great pace with sensors and deep learning, but there is limited public confidence because of limited comparative proof against human motorists. The current paper creates a hybrid deep learning model by linking traffic signs and object perception with convolutional neural networks (CNNs) alongside temporal signal perception with recurrent (RNN/LSTM) networks. Conditional imitation learning facilitates contextual decision-making under changing conditions of roads, traffic, and weather. Training is assisted by vast datasets like GTSRB, Comma.ai, and BDD100K, pre-processed by augmentation along with fusion of camera, LiDAR, and radar signals. 95% of validation accuracy and virtually flawless (99%) traffic sign compliance are attained, surpassing human motorists (91%). Comparative analysis shows averaged reaction time of 0.32 s against 1.25 s, averaged lane deviation of 5 cm against 12 cm, and substantially reduced abrupt braking occurrences (3 per 100 km against 11). The findings demonstrate the model's quicker reaction, higher accuracy, and more cautious driving. In pursuit of transparency, explainable AI techniques (attention maps, SHAP values) are included, enhancing interpretability and confidence. It gives empirical proof that AVs can reliably surpass human-driven vehicles in major measures of safety, lending support to AVs being eventually permitted in widespread real-world transportation.

Keywords: Autonomous Vehicles, Manual Driving, Sensor Data, Road Safety, Deep Learning, Traffic Sign Recognition, Human-Computer Comparison, Driving Behaviour, CNN-LSTM

Abstract: Agriculture faces pressing challenges such as climate change, water scarcity, and crop diseases, demanding efficient and sustainable solutions. This research presents a smart farming system that integrates IoT sensors, edge computing, and artificial intelligence for real-time monitoring and decision-making. Environmental factors like soil moisture, temperature, humidity, and pH are continuously tracked, while on-site crop leaf images enable disease detection. A deep learning model deployed on edge devices ensures offline functionality, reducing reliance on internet connectivity. The system transitions from MobileNetV2 to the more accurate EfficientNetB3 architecture, achieving improved performance without sacrificing efficiency. This integration enhances productivity, optimizes water usage, and supports timely interventions. Designed for scalability and cost-effectiveness, it offers practical benefits to small and mid-scale farmers. By merging AI with IoT-based sensing, the approach transforms traditional agriculture into a smarter, more resilient practice.

Keywords: Internet of Things (IoT), Edge Artificial Intelligence (Edge AI), MobileNetV2, Convolutional Neural Networks (CNN), Raspberry Pi

Abstract: In today's digital landscape, the detection and filtering of unwanted communications, known as spam, are an integral part of protecting cyber security and trust in users. This paper presents an AI spam detection system that uses state-of-the-art machine learning (ML) and natural language processing (NLP) methods to identify and filter bad or irrelevant online messages. The system analyzes text patterns, frequency of suspicious words, and sender information. We performed a comparative study with three classifiers, Naive Bayes, Support Vector Machine (SVM), and a Neural Network model, to differentiate spam and valid messaging content. The models are trained on large labeled datasets and show good accuracy for classifying text and identifying various threats such as phishing attacks, online scams, and unsolicited marketing messages. Artificial intelligence can be applied to improve spam filtering in real-time, and is a scalable and intelligent method to the difficult problems in digital communication today

Keywords: AI-driven spam detection framework, cyber security, Machine learning, Natural language processing, Naive Bayes, Support Vector Machine, Neural Network model.

Abstract: Parking space management has become extremely difficult due to the sharp rise in urban vehicle ownership, which has increased carbon emissions, wasted fuel, and traffic congestion. Urban mobility is inefficient due to the lack of real-time monitoring, intelligent allocation, and predictive capabilities in traditional parking management systems. In order to maximize parking utilization, this paper introduces a Smart Parking Management System (SPMS) that combines artificial intelligence (AI) and Internet of Things (IoT) sensors. IoTenabled sensors identify the presence of vehicles and send real-time occupancy data, while AI-based algorithms evaluate both live and historical data to forecast availability and direct drivers. digital payments, booking, and navigation are all made possible by a mobile application. Reduced cruising time, better space use, lower emissions, and scalable deployment for smart cities are the goals of the suggested system. Architecture, implementation options, testing methods, assessment metrics, and future directions are all covered in the paper.

Abstract: The Internet of Things (IoT) has emerged as a transformative technology in healthcare, enabling real-time monitoring, early detection, and improved patient care through interconnected sensors, wearables, and cloud platforms. This paper presents a unified study of IoT applications in healthcare monitoring systems with a focus on chronic diseases. Specifically, it synthesizes research on non-invasive blood glucose monitoring (GluQo), cloud-assisted asthma monitoring, and multi-parameter healthcare frameworks. The paper highlights IoT architectures, system designs, and implementation results, while also addressing challenges of data privacy, interoperability, and scalability. Future directions include AI-driven IoT frameworks, blockchain-enabled security, and 5G-enabled telehealth.

Keywords: IoT, Healthcare Monitoring, Glucose Monitoring, Asthma Monitoring, Cloud Computing

Abstract: This study uses a carefully chosen patient dataset that includes a variety of demographic traits, lifestyle factors, and medical histories to reliably predict heart illness using logistic regression. A representative portion of the information is used to train the model (Logistic Regression), which was selected due to its efficacy in binary classification, to find intricate patterns that may indicate the risk of heart illness. A comprehensive health profile that includes lifestyle variables, physiological markers, and patient demographics allows for a more nuanced risk assessment. Extensive testing on an independent sample confirms the model's excellent discrimination accuracy between those with and without heart disease. This study advances data-driven healthcare by demonstrating how Logistic Regression might improve the precision of heart disease prediction. The findings have implications for proactive cardiovascular health management and individualized patient care through educated clinical decision-making.

Keywords: F1 score, Heart Disease, Logistic Regression, Precision, Recall, Sensitivity Analysis, Variable Selection.

Abstract: Real-time healthcare monitoring uses wearables and bedside sensors to watch patients’ vital signs and alert caregivers quickly. Sending all data to the cloud can be slow and risky for privacy. Edge computing processes data close to where it is collected, which lowers delay and saves bandwidth. Federated learning lets many devices train a shared model without sending raw patient data, which supports privacy. This paper presents a simple, practical framework that combines edge computing and federated learning for faster, safer health monitoring. Our design chooses which devices should join each training round based on their battery, signal quality, and recent data. We reduce network load using light model updates with quantization and sparsification, and we add secure aggregation and differential privacy to protect patients’ information. We also include small “personalization” parts in the model so each device can adapt to its patient. We describe a step-by-step method, an objective that balances accuracy, latency, and energy, and an evaluation plan using public physiological datasets under changing network conditions. Expected results show similar accuracy to standard training, with lower latency, fewer false alarms, and less bandwidth use. This work offers a clear path to deploy trustworthy, real-time monitoring at the edge.

Keywords: Edge Computing, Federated Learning, Healthcare Monitoring, Wearable Devices, Resource Optimization, Medical IoT

Abstract: Artificial Intelligence (AI) is transforming assistive technologies into intelligent systems that enhance accessibility across visual, auditory, motor, and cognitive domains. This review examined 52 works published between 2023 and 2025, with 40 peer-reviewed studies systematically analyzed using PRISMA guidelines. The Accessibility Impact Score (AIS) was introduced as a novel framework to evaluate usability and effectiveness. Findings show that AI-powered tools such as smart glasses, adaptive exoskeletons, and multimodal learning platforms outperform traditional assistive devices. Visual and motor applications achieved the highest AIS values, while auditory and cognitive tools demonstrated strong emerging potential. The integration of multimodal AI, including voice, vision, haptics, and brain–computer interfaces, enables proactive and context-aware support. These results highlight AI’s role in enhancing independence, social participation, and quality of life. The review also emphasizes the importance of open datasets for reproducibility and the need for ethical, inclusive, and scalable adoption of AI in accessibility. Overall, AI offers a paradigm shift toward inclusive, human-centered assistive systems with potential applications in healthcare, education, and daily living.

Keywords: Artificial intelligence, Assistive technology, Computer vision, Large language models, Machine learning, Natural language processing.

Abstract: A Comparative Study of Sentimental analysis using the NLP and different machine learning and deep learning techniques focuses on analyzing the subjective information conveyed with the expression. This encompasses appraisals, opinions, attitudes or emotions towards a particular subject, individual, or entity. Conventional sentiment analysis solely consider text modalities and the derives sentiment by identifying the semantic relationships between the words in the sentences. Despite this, some expressions, such as exaggeration, sarcasm and humor pose a challenge for the automated detections when conveyed only through the texts. The similar approach can precisely determine the implied sentiment polarities which contain all positive, neutral, Negative sentiment. This research communities can shown significant interest with the topic because of its potential for both the practical application and education related research. With this fact, the paper aims to present all analysis of recent ground breaking research studies which helps the deep learning models in many modalities and works.

Keywords: Sentiment Analysis, Deep Learning, Emotion Recognition, Opinion Mining, Sarcasm and Humor Detection.

Abstract: The quality of images captured by budget-friendly smartphones degrades significantly in low-light conditions due to hardware limitations. To resolve this, we present a lightweight, end-to-end deep learning framework designed to function as a software-based Image Signal Processor (ISP) for on-device enhancement. Our approach is centered on a U-Net architecture, trained on a hybrid dataset combining specialized low-light pairs (LoL) and general high-quality photographs (MIT-Adobe FiveK) to ensure robust and aesthetically pleasing results. The model, which contains only 2.90 million parameters, is optimized using a composite loss function balancing pixel-wise accuracy and structural integrity. Quantitative evaluation shows our model achieves a highly competitive PSNR of 17.24 dB on the LoL Dataset. A key finding from our ablation studies reveals that for a network of this scale, a simpler architecture without residual connections performs marginally better, providing a valuable insight for future lightweight model design. Overall, our work demonstrates a superior trade-off between performance and computational efficiency, establishing a promising foundation for bringing superior photographic computation on a variety of mobile devices.

Keywords: Deep Learning, U-Net, Mobile ISP, CNN, Lightweight Neural Networks, On-Device AI, Computational Photography, Edge Computing, Low-Light Image Enhancement.

Abstract: In the fast growing industry of the Artificial Intelligence and Machine Learning Applications, the face recognition also plays an important role in many sector including the one used in for the identification of the person.The Face Recognition system can be used in schools, colleges, offices etc. for marking of hassle free and fast attendance so that the teaching faculty gets time to focus on delivering the content rather than taking attendance. time consuming and highly prone to errors. The results indicate that the face recognition system is a viable and effective solution, significantly reducing the errors and saving time.

Abstract: OCT imaging is an essential test to help diagnose retinal diseases involving Choroidal Neovascularization (CNV), DME, Drusens, etc. They get accurately and automatically classified with the early help of ophthalmologists for diagnosis and treatment. In semi-supervised learning practice, active learning is a popular technique specialized in the selection of training sets. The OCT dataset comprises a total of 83,500 retinal images of high quality. In a similar manner, the dataset assigns an equal quantity of images to all four classes. We resize the images to make them 224 × 224 pixels. Normalize the image and apply some augmentations. Improvements used on the images include flipping, rotating, zooming and changes in the brightness. Improvements assist in reinforcing our model by avoiding overfitting. The adam optimizer is used to train the model while learning specific retina characteristics with the categorical cross-entropy loss function. The suggested approach attained an overall accuracy of 95.2% with all class precision, recall and F1-score being 94%. The Grad-CAM visualisation shows the model focusing correctly on the retina. The study suggested that deep learning techniques or explainable AI can help ophthalmologists diagnose retinal diseases automatically, which can further help in clinical decision making. The statement reveals that the transfer learning models can provide reliable as well as explainable results in retinal diseases.

Keywords: OCT imaging, learning, retinal, overfitting and optimizer.

Abstract: Cyber crime and cyber security remain central issues for governments, businesses, and individuals in the 21st century. This paper provides a comprehensive review of the contemporary landscape of cyber crime, examines the major threat vectors and actors, and evaluates the technical, organizational, and policy responses used to mitigate risk. Combining a critical literature review, case-study analysis, and proposed methodological approaches for empirical investigation, the paper identifies trends such as the commercialization and specialization of cybercrime-as-a-service (CaaS), the growing sophistication of state-sponsored operations, and the persistent vulnerabilities arising from human factors and legacy systems. The discussion synthesizes findings to produce actionable recommendations for practitioners and policymakers, including adoption of layered defense strategies, improved incident response and forensic readiness, public–private collaboration, and regulatory harmonization. Limitations and directions for future research are outlined.

Keywords: cyber crime, cyber security, cyber threat intelligence, incident response, digital forensics, policy

Abstract: This paper examines algorithmic bias in AI systems used for military decision-making, identifies key sources of unfairness, and demonstrates practical mitigation strategies with implemented machine-learning experiments. We generate a synthetic but realistic dataset that mimics decisions (e.g., target identification / threat classification) with a binary sensitive attribute (e.g., group A vs group B). We implement baseline classifiers (Logistic Regression, Random Forest), measure fairness-related metrics (statistical parity difference, equal opportunity difference, disparate impact), and apply two mitigation strategies: reweighing (pre-processing) and group-specific thresholding (post-processing). Results include accuracy, fairness trade-offs, and visualizations. The paper ends with recommendations and limitations. Keyword: Algorithmic bias, fairness, military AI, reweighing, thresholding, fairness metrics, machine learning.

Abstract: In today's fast-paced world, individuals often rely on static daily planners to manage their tasks. These planners are designed based on fixed priorities, deadlines, and time slots, but they rarely account for the emotional and mental well-being of the individual using them. When emotions are not considered, users may find themselves overwhelmed, unmotivated, or mentally unprepared for the tasks scheduled for them, which ultimately reduces productivity and impacts mental health. EmotiPlan is an innovative AI-powered emotion-based day planner that detects the user's current emotional state through facial recognition and creates a dynamic, personalized schedule accordingly. It uses modern technologies like DeepFace and OpenCV to analyze facial expressions and interprets the user's mood in real-time. Based on the detected emotion— whether it's happiness, sadness, anger, or fatigue—the system adjusts task priorities and suggests an optimal timetable that balances productivity with well-being. Built on the MERN stack with a Python-based Flask API, EmotiPlan integrates backend scheduling logic with frontend user interaction, while also syncing events with Google Calendar. The first phase of the project establishes a working prototype capable of detecting mood, mapping it to activity types, and presenting a visual schedule to the user through an intuitive dashboard interface.

Keywords: Facial Emotion Recognition, MERN Stack, Deep Face, AI Planner, Human-Centered Productivity.

Abstract: In the contemporary world, the rapid pace of technological advancement and the increasing diversity of career options have created a significant challenge for students and professionals seeking guidance in making informed career decisions. Traditional career counseling methods are often limited in scope, dependent on subjective assessments, and lack the ability to provide personalized, data-driven recommendations at scale. This research presents an AI Career Guidance Platform designed to address these limitations by leveraging artificial intelligence, machine learning algorithms, and predictive analytics to provide comprehensive, personalized career guidance.

Abstract: Emotions play a vital role in human communication, influencing decision-making, social interaction, and overall well-being. The ability to automatically recognize emotions across multiple modalities has become increasingly important in fields such as mental health monitoring, intelligent customer support, and affective computing. This work proposes a Multimodal Emotion Recognition System that integrates speech, text, and facial expression analysis to achieve a more reliable and comprehensive understanding of human emotions. For speech, features such as Mel Frequency Cepstral Coefficients (MFCCs) and pitch are extracted and analyzed using Random Forest classifiers along with Deep Neural Networks (DNNs) for improved performance. Text-based emotion recognition leverages the contextual learning capabilities of Recurrent Neural Networks (RNNs) and Long Short-Term Memory (LSTM) models to capture linguistic nuances. Facial expression recognition is conducted using Convolutional Neural Networks (CNNs), enhanced with wavelet transforms for better feature representation. The fusion of these modalities helps address the limitations of single-source emotion detection, leading to more accurate and holistic recognition. The proposed system is deployed as a user-friendly web application built with Flask, HTML, and CSS, making it accessible for practical use. This research contributes to the advancement of multimodal affective computing and highlights the potential of integrated ML and DL approaches for real-world emotion-aware applications.

Abstract: Rainfall prediction plays a critical role in agriculture, water resource management, and disaster preparedness. This project focuses on predicting rainfall in Australia using a comprehensive weather dataset containing meteorological variables such as temperature, humidity, wind speed, pressure, and historical rainfall records. The research problem addressed in this study is the uncertainty and inaccuracy of traditional forecasting methods, which often fail to capture complex, non-linear weather patterns. The main objectives of this study were to preprocess the dataset, handle missing values, apply data balancing techniques using SMOTE, and implement machine learning models for accurate rainfall forecasting. Various classification algorithms were applied, including Logistic Regression, Decision Tree, Random Forest, and Gradient Boosting. The models were evaluated using accuracy, precision, recall, and F1-score. The findings indicate that ensemble methods performed significantly better compared to simple classifiers. Among all models, Random Forest achieved the highest accuracy of approximately 85%, with humidity, temperature, and wind-related features emerging as the most influential predictors

Abstract: Customer service operations across industries face mounting challenges including high operational costs, inconsistent service quality, limited availability, and scalability constraints during peak demand periods. Traditional customer support systems rely heavily on human agents, resulting in longer response times, higher labor costs, and potential for human error or bias in service delivery. An AI Based Chatbot system provides a transformative solution by leveraging advanced natural language processing techniques, machine learning algorithms, and conversational AI technologies to deliver automated, intelligent, and personalized customer interactions. This intelligent system processes natural language queries, understands user intent, maintains conversation context, and provides accurate responses through sophisticated language models and knowledge base integration. By implementing cutting-edge technologies including BERT transformers, intent classification algorithms, named entity recognition, and dialogue management systems, the chatbot achieves human-like conversation capabilities while maintaining consistency and accuracy across all interactions. The AI-based approach enables 24/7 availability, instant response times, multilanguage support, and seamless escalation to human agents when necessary. With features including sentiment analysis, personalized recommendations, conversation history tracking, and continuous learning capabilities, this system represents a significant advancement in customer service automation while reducing operational costs and improving user satisfaction. The implementation demonstrates substantial improvements in response accuracy, conversation flow management, and adaptation to diverse customer queries and contexts.

Keywords: AI Chatbot, Natural Language Processing, Conversational AI, Customer Service, Machine Learning, Intent Recognition, BERT, Dialogue Management.

Abstract: The proliferation of digital media and social networking platforms has led to an unprecedented spread of misinformation and fake news, posing significant threats to democratic processes, public health, and social stability. Traditional methods of news verification rely heavily on manual fact-checking, which is time-consuming, resource-intensive, and unable to scale with the exponential growth of online content. An AI Based Fake News Detection system provides a revolutionary solution by leveraging advanced machine learning algorithms, natural language processing techniques, and deep learning models to automatically identify and classify news articles as authentic or fabricated. This intelligent system analyzes multiple features including textual content, linguistic patterns, source credibility, social media engagement metrics, and metadata to make accurate predictions about news authenticity. By implementing sophisticated algorithms such as Support Vector Machines, Random Forest, Long ShortTerm Memory networks, and transformer-based models like BERT, the system achieves high accuracy in distinguishing between legitimate journalism and deliberately misleading content. The AIbased approach not only provides real-time detection capabilities but also offers scalable solutions that can process millions of articles simultaneously across multiple languages and platforms. With features including sentiment analysis, source verification, fact-checking integration, and confidence scoring, this system represents a significant advancement in combating misinformation while supporting informed decision-making in the digital age. The implementation demonstrates substantial improvements in detection accuracy, processing speed, and adaptability to emerging fake news patterns.

Keywords: Fake News Detection, Machine Learning, Natural Language Processing, Artificial Intelligence, Deep Learning, Misinformation, BERT, Classification

Abstract: Artificial Intelligence (AI) has rapidly emerged as a transformative technology that influences multiple sectors including healthcare, education, business, and research. Its fast growth has generated diverse reactions within the AI community, ranging from optimism to concern. On one side, AI is welcomed for its ability to enhance productivity, improve decision-making, and drive innovation. On the other hand, critical voices raise concerns about ethical challenges, privacy risks, job displacement, and algorithmic bias. This study examines these mixed reactions by analyzing both positive and negative perspectives of researchers, professionals, and policymakers. The findings suggest that while the AI community largely acknowledges the revolutionary potential of AI, it also emphasizes the urgent need for responsible development, ethical guidelines, and transparent governance. Overall, the reaction of the AI community reflects a balanced view that combines excitement for opportunities with caution about risks, ensuring that AI serves humanity in a safe and beneficial way.

Abstract: In today’s fast-evolving academic and professional landscape, students and graduates face numerous challenges when selecting the right career path. Traditional counselling methods are often generic and unable to cater to individual aspirations, skills, and interests. This paper presents the development of a “Smart Career Counsellor” — an AI-driven system designed to provide personalized career guidance using data analytics and machine learning. The system collects data from users through interactive questionnaires assessing academic performance, skills, and personal preferences. Based on this information, it generates tailored career recommendations and learning paths. The proposed model integrates Artificial Intelligence algorithms, Natural Language Processing, and predictive analytics to enhance accuracy and adaptability. The study aims to bridge the gap between education and employment by providing data-driven insights for informed decision-making. The Smart Career Counsellor demonstrates how AI can transform career guidance into a more efficient, objective, and personalized experience for students and graduates, ultimately supporting their professional success in a technology-driven world.

Keywords: Artificial Intelligence, Career Counselling, Machine Learning, Personalized Guidance, Student Career Planning, Data Analytics, Smart System, Predictive Analysis, Educational Technology, NLP-based Recommendation

Abstract: Skin cancer is a major global health burden, and early detection markedly improves outcomes. Yet many patients face delayed diagnosis because specialist dermatology expertise is scarce or unevenly distributed, especially in underserved regions. We propose an AI-driven decision-support system that analyzes clinical and dermoscopic images to flag suspicious lesions for clinician review. Trained on large, curated image datasets, the model learns visual patterns linked to malignancy, analogous to experiential learning in clinical practice. In reader studies, deep learning systems have achieved dermatologist-level performance and, when used alongside clinicians, can enhance diagnostic accuracy and triage efficiency. Integrated responsibly into workflows, such tools may expand screening reach, shorten time to specialist assessment, and enable earlier intervention while complementing—not replacing—clinical judgment.

Keywords: Artificial Intelligence, Skin Cancer, Image Analysis, Deep Learning, Dermatoscopy, Diagnostic Tool

Abstract: As a reply to the tremendous complexity and extent of modern agricultural weed control, a statistical analytical review is being presented in this study of all modern scholarly contributions in the area of deep learning (DL), optimization algorithms, ecological control systems, and biochemical valorisation approaches. Its main objective is to bundle these disparate methods into a single benchmarking framework on six performance criteria: Weed Detection Accuracy, Precision, Reliability, Time Complexity, Memory Complexity, and Makespan Sets. Therefore, methodologies such as DC-YOLO, Efficient Net-based transfer learning, RSA-enhanced YOLOv3, SCR-DETR, and RCNN are compared in this review with approaches already using microbial weed suppression models, anaerobic bioreactors, and graph-based simulation techniques. Each of the above approaches is being evaluated both qualitatively and quantitatively while data were then gleaned from reported empirical findings or approximated through interpolation of cross-study performance. In the comparative matrix established through the dataset, detection efficiency, scalability, resource requirements, and robustness are displayed across a wide range of use cases; from UAV surveillance, thermal classifications, to mechanical weeding. All vision-based DL approaches score detection accuracies well above 93%, but the cost is increased memory usage and time complexity incurred when the complexity of the scene increases in process. Thus, this review identifies research gaps in data fusion, model interpretability, and real-world deployment, while proposing future integration pathways such as eco-AI coupling and secure visual processing for different scenarios. This work lays the groundwork for precision agriculture as an analytical resource in weed control events at some point from future technological applications in robust adaptive and environmentally sustainable weed control treatments in process.

Keywords: Weed Detection, Deep Learning, Precision Agriculture, Performance Benchmarking, Image Segmentation, Scenarios.

Abstract: Cardiovascular diseases (CVDs) are the leading cause of global mortality, necessitating early and accurate detection methods to improve patient outcomes. Traditional diagnostic approaches, such as ECGs and angiograms, are often invasive, costly, or require specialized expertise, making non-invasive alternatives highly desirable. Recent advancements in artificial intelligence (AI) and machine learning (ML) have enabled the analysis of retinal images for heart disease prediction, leveraging the structural and functional similarities between retinal vasculature and coronary arteries. Retinal imaging techniques, such as fundus photography and optical coherence tomography (OCT), allow for non-invasive visualization of microvascular changes linked to cardiovascular conditions. ML models, including convolutional neural networks (CNNs) and hybrid deep learning architectures, can effectively analyze these images to detect abnormalities indicative of heart disease. This review explores various datasets, feature extraction methods, and classification techniques used in retinal image analysis for cardiovascular risk assessment, comparing their effectiveness in predictive modelling. Despite promising advancements, challenges such as data availability, model generalizability, explainability, and clinical integration remain critical. Future research should focus on developing robust, interpretable AI models, enhancing dataset quality, and addressing real-world implementation barriers to establish retinal imaging as a reliable tool for early heart disease detection.

Keywords: Heart Disease, Retinal Imaging, Machine Learning, Deep Learning, Cardiovascular Disease, Medical Image Processing.

Abstract: In today’s well-being-conscious era, individuals strive to maintain balanced nutrition and monitor their dietary intake, yet manual tracking methods remain cumbersome and prone to inaccuracies. The proposed Artificial Intelligence Driven AI-Powered Nutrition Evaluation System provides a smart and efficient solution to analyze and evaluate the nutritional composition of food items using artificial intelligence and image recognition. By capturing or uploading a meal image, the framework identifies the food components and estimates their nutritional values, including calories, proteins, carbohydrates, fats, and essential micronutrients. The framework employs Convolutional Neural Networks (CNNs) for image classification and integrates a nutritional database for value computation. It further customizes recommendations based on user-specific conditions such as diabetes, obesity, or deficiencies, assisting in well-beingy decision-making. The AI-driven approach significantly minimizes human error, enhances user engagement, and promotes sustainable well-being monitoring through automation and personalization.

Keywords: Artificial Intelligence, Nutrition Tracking, Machine Learning, CNN, Health Monitoring, Image Recognition.

Abstract: Road Safety and public health are two critical domains where simple preventive measures such as wearing helmets and face mask while riding plays a major role in reducing accidents and disease transmission. manual monitoring of such compliance is difficult to scale, especially in densely populated areas, making automated solutions is necessary. Recent advances in computer vision and deep learning have enabled intelligent surveillance systems capable of detecting helmet and mask violations in real time. Techniques such as YOLO (You Only Look Once), Convolutional Neural Networks (CNNs), and Optical Character Recognition (OCR) have shown strong performance for multi class detection tasks, including helmets, masks, and license plates. This review Systematically explores research contributions in three categories: helmet detection, mask detection, and combined helmet + mask detection. By comparing traditional machine learning approaches with state-of-the-art deep learning frameworks, the paper highlights the growing potential of unified AI-powered systems for improving public safety and traffic enforcement.

Keywords: Road Safety, Public Safety, Intelligent Surveillance, Helmet Detection, Face Mask Detection, YOLO, OCR.

Abstract: In the last few years, fake customer reviews have become a big problem for companies and consumers in online shopping websites. The history of this project started when many shopkeepers complained about fake negative reviews hurting their business. When customers buy products based on fake reviews, they feel cheated and loose trust in online shopping platforms. This problem statement shows the need to find better ways to find fake reviews and remove them from websites. The solution is to use AI technology like BERT models that can understand language patterns in reviews and identify which ones are fake. By combining BERT with other methods like CNN and capsule networks, the accuracy of detecting fake reviews improves a lot. The system will look at things like writing style, emotional words, and unusual patterns that might show fake reviews. Tests showed that our method finds fake reviews with 92% accuracy which is better than older methods. This project will help make online shopping more trustworthy for everyone and protect honest businesses from getting bad reviews that are not real.

Keywords: Fake reviews detection, BERT models, sentiment analysis, customer feedback, machine learning, neural networks, e-commerce platforms, text classification, data augmentation, transformer models

Abstract: In the modern era, Reinforcement Learning (RL) has evolved from foundational experiments in classic control tasks to sophisticated systems facing contemporary challenges. While early tasks featured discrete action spaces and observable states, modern problems often involve continuous control and complex dynamics. This progression has created a significant algorithmic gap, requiring different approaches for optimal performance. This paper presents a comparative analysis to characterize this gap by benchmarking two influential algorithms on representative tasks: the value-based Deep Q-Networks (DQN) on a discrete control problem, and the policy-gradient Proximal Policy Optimization (PPO) on a continuous control problem. The analysis reveals the specialized strengths of each method, demonstrating that DQN achieves high performance in its intended domain, while PPO's architecture is well-suited to the stability requirements of more complex, continuous environments. These findings provide an empirical basis for understanding the distinct capabilities of these algorithmic classes, clarifying their respective domains of application and highlighting the importance of matching algorithmic design to problem complexity.

Keywords: Reinforcement Learning, Comparative Analysis, Deep Q-Networks (DQN), Proximal Policy Optimization (PPO), Algorithmic Gap.

Abstract: Crops are always in demand in the country, not only for the lives of the people, but also for eco nomic growth, so growing crops is of utmost importance. Using standard technology also increases efficiency and lessens the workload of the farmers. Therefore, in order to increase productivity, it is important to know about soil moisture and types of crops. Each variety of crop and the associated soil requires a particular amount of water, so the project need to make the most of what is available. In order to achieve this, it must utilize modern technology and tools. This paper focuses on an automated irrigation system, i.e., irrigating fields only when they need to be watered, by utilizing machine learning algorithms. Real-time readings of soil moisture, fertility, and pH are sensed through sensors and are available on the system.

Keywords: Agricultural yield prediction, Crop yield forecasting, Machine learning in agriculture, Regression models in agriculture

Abstract: AbhyasX - Smart Learning Ecosystem is a digital learning platform primarily designed for engineering students to provide structured and accessible study resources. The platform offers free subject-wise notes across all branches along with premium courses that include advanced learning materials, career-oriented guidance, and practice resources. With secure authentication (Clerk), database management (Supabase), and payment integration (Razorpay), the system ensures reliability, safety, and smooth user experience. Students are also provided with personalized dashboards and progress monitoring features to help them track their academic journey effectively. Built using modern web technologies such as Next.js and Tailwind CSS, and supported through scalable REST APIs, the system is designed for future growth and flexibility. Planned future enhancements include an interactive quiz module for each subject to strengthen conceptual learning and self-assessment. Additionally, the platform will expand its scope by including resources for MCA (Master of Computer Applications) students, making it a comprehensive learning ecosystem not just for engineering, but also for postgraduate learners. With its structured design, advanced technologies, and continuous feature expansion, AbhyasX aims to evolve into a reliable, student-centric Smart Learning Ecosystem that empowers learners with accessible, personalized, and effective digital education.

Keywords: AbhyasX, Smart Learning Ecosystem, Personalized Learning, Engineering Education, MCA Education, Digital Platform, Next.js, Supabase, Clerk Authentication, Razorpay Integration, E-Learning, Student Dashboard, Progress Tracking, Adaptive Learning, Online Education.

Abstract: Poly cystic Ovary Disorder is one of the most prevalent endocrine conditions affecting women of reproductive age, yet its clinical presentation varies widely across reproductive metabolic and hormonal domains. Current diagnostic frameworks and computational models often simplify PCOD into a binary classification present or absent ignoring its intrinsic heterogeneity. This oversimplification hinders precision medicine and the development of tailored treatment plans. This paper introduces a novel framework that combines self-supervised learning (SSL) and deep clustering to uncover hidden PCOD phenotype directly from unlabeled clinical records. Contrastive SSL approaches such as Sim CLR and MoCo are used to learn patient embeddings which are subsequently clustered using Deep Cluster and SwAV. The resulting clusters are validated against clinical indicators including body mass index (BMI) hormonal ratios and insulin resistance measures. Experiments revealed three distinct clusters: mild PCOD with near normal parameters moderate PCOD with hormonal irregularities and severe PCOD with metabolic risks. The framework was deployed in a Django based clinical platform providing real time cluster assignments visual analytics and patient level reports. By moving beyond binary classification this work demonstrates the potential of SSL driven phenomapping to enable precision gynecology supporting individualized treatment strategies and advancing scalable clinical decision support systems.

Keywords: Polycystic Ovary Disorder, self-supervised learning, contrastive learning deep clustering, clinical phenotyping, decision support.

Abstract: Vehicle theft continues to be a major global concern, demanding advanced and intelligent security systems capable of real-time monitoring and immediate threat response. This paper presents the design and implementation of a smart Vehicle Theft Detection System (VTDS) that integrates Arduino UNO, GPS (NEO-6M), and GSM (SIM800L) modules with multiple sensors to detect, track, and respond to unauthorized vehicle activities. The system utilizes a vibration sensor, ignition status monitor, and accelerometer (MPU6050) to identify intrusion attempts, ignition tampering, or abnormal vehicle motion. Upon detection of a threat, the system instantly sends an SMS alert containing GPS coordinates and a Google Maps link to the registered user, enabling accurate real-time tracking. Additionally, the vehicle engine can be remotely disabled via SMS, while a buzzer alarm provides immediate local notification. The system was successfully tested under various simulated theft conditions, showing high accuracy, minimal false triggers, and reliable communication with an average alert latency below four seconds. This work demonstrates an affordable, scalable, and energy-efficient IoT-based vehicle security solution that enhances safety, reduces theft risk, and enables rapid recovery in case of unauthorized access.

Keywords: Vehicle theft detection, Arduino UNO, GPS tracking, GSM communication, IoT security system, accelerometer sensor, ignition monitoring, vibration detection, real-time alert, smart vehicle system.