Current Issue
Latest published peer-reviewed scholarly articles in International Journal of Research, Exploration & Knowledge (IJREK)
Volume 1, Issue 1 (October 2026)
Inaugural Issue of International Journal of Research, Exploration & Knowledge (IJREK)
Articles in this Issue (105)
Design and Evaluation of a Mobile-Friendly College Information
College websites are important sources of information about courses, examinations, notices, contact details, and student services. However, a website that is difficult to navigate on a small screen or that does not follow accessibility principles can create barriers for students, including people who use assistive technologies. This study proposes the design and evaluation of a mobile-friendly college information portal using usability testing and accessibility checks. The objective is to determine whether a redesigned interface helps users complete common information-seeking tasks more successfully and with fewer errors, while also improving conformance with selected accessibility practices. A lightweight prototype will be developed using HTML, CSS, and JavaScript, with responsive layouts and accessible navigation, headings, labels, contrast, and form controls. The evaluation will compare an initial interface with an improved version using predefined tasks such as locating course information, finding an examination notice, and submitting an enquiry. Measures will include task completion rate, task time, user errors, automated accessibility findings, and perceived usability measured with the System Usability Scale. At the time of this draft, implementation and user testing have not been completed; therefore, no empirical results are claimed. The final paper will report observed measurements, discuss limitations, and identify practical design recommendations based on the evidence collected. The intended contribution is a reproducible, low-cost evaluation procedure for improving the usability and accessibility of college information portals.
Predictive Mechine maintenace using AI&ML
Unexpected machine failures can interrupt production, increase repair expenses, and create safety risks. Predictive machine maintenance uses sensor data, Artificial Intelligence (AI), and Machine Learning (ML) to estimate when equipment may develop a fault. Instead of servicing every machine at fixed intervals or waiting for a breakdown, the system continuously monitors parameters such as vibration, temperature, pressure, current, and operating hours. Data is cleaned and analysed, and a trained ML model identifies abnormal behaviour or predicts the remaining useful life of a component. The system can then generate an alert so that maintenance is planned before failure occurs. This paper explains the concept, working process, major technologies, model design, applications, benefits, limitations, and a practical implementation framework. It also discusses data quality, model evaluation, cybersecurity, and human verification. Predictive maintenance can improve equipment availability and maintenance planning, but its results depend on representative data, suitable sensors, and regular model monitoring.
User experience and trust in AI powered virtual assistant
AI-powered virtual assistants are increasingly used for information retrieval, task support, customer service, and everyday digital interaction. Their usefulness depends not only on technical performance but also on the experience they provide and the trust users place in their responses. This paper presents a conceptual study of user experience and trust in AI-powered virtual assistants. It examines factors such as perceived usefulness, ease of use, response quality, transparency, privacy, reliability, and the handling of errors. A proposed research framework connects these factors with user satisfaction, trust, and intention to continue using an assistant. The paper outlines a survey-based methodology using a structured questionnaire and proposes descriptive and relationship-based analysis. As this is a proposed study, no participant data or statistical findings are claimed. The framework highlights the importance of clear communication, privacy protection, consistent performance, and appropriate disclosure of AI limitations for developing trustworthy virtual assistants.
Explainable Machine Learning-Based Student Academic Performance Prediction
Student academic performance prediction can help educational institutions identify learners who may benefit from timely, targeted support. Machine-learning models can combine carefully selected indicators such as prior assessment results, attendance, assignment completion and learning-platform engagement to estimate likely performance. However, a prediction that cannot be explained may be difficult for teachers and students to trust, and may conceal data quality problems or unfair patterns. This paper presents a conceptual framework for explainable machine-learning-based student academic performance prediction. The proposed workflow includes data governance, preprocessing, feature engineering, baseline and advanced model comparison, interpretable explanations, fairness checks, and human-led intervention. Candidate models include logistic regression, decision trees, random forests and gradient-boosted trees, with SHAP- or permutation-based explanations where appropriate. Evaluation should use held-out data, class-sensitive metrics, calibration and subgroup analysis rather than accuracy alone. The framework is designed to support—not replace—professional educational judgement. No original dataset or experimental result is claimed; model performance must be established through a documented local study.
AI-Powered Phishing Detection
Phishing is a social-engineering threat in which attackers impersonate trusted organisations or individuals to persuade users to disclose information, open harmful attachments, or visit deceptive websites. Conventional filters can be challenged by newly created domains, changing message patterns, and carefully crafted text. This paper proposes an AI-Powered Phishing Detection system that analyses selected email, URL, and message features to estimate whether an item may be suspicious. The workflow includes input validation, feature extraction, machine-learning classification, risk scoring, explanation of influential indicators, and a user-facing warning or review action. Candidate features include URL structure, domain signals, message wording, sender metadata, and selected email-header indicators where available. The system is a defensive screening aid, not a guarantee that a message is safe or malicious. The paper describes architecture, workflow, implementation, evaluation, applications, and limitations. Diagrams and chart values are illustrative, not experimental results.
Explainable AI for Early Prediction of Student Academic Risk Using Machine Learning
Student academic risk prediction aims to identify learners who may need timely academic support before final outcomes are known. Machine learning can analyse historical indicators such as attendance, assessment performance, assignment completion, and learning-platform engagement to estimate the likelihood of academic difficulty. However, a prediction is more useful to educators when its basis can be understood. This paper proposes an explainable machine-learning framework that combines early risk estimation with interpretable explanations for students, faculty, and academic advisors. The workflow includes data governance, preprocessing, model training, validation, risk scoring, local and global explanations, and human review. Candidate models include logistic regression and decision trees as interpretable baselines, alongside more complex models where justified. Explanation methods such as feature importance and SHAP-style contribution analysis can help communicate which recorded indicators influenced a prediction. The system is designed to support—not replace—professional judgement, and it should not automatically penalise students. Fairness, privacy, data quality, and clear communication are treated as core design requirements. The proposed architecture and evaluation plan are presented for an academic prototype; no experimental performance results are claimed.
AI Based Email security System
Email remains a major communication channel for individuals and organizations, but it is frequently abused for phishing, spam, malware delivery, impersonation, and social engineering. Traditional filtering methods often rely on fixed rules, known signatures, or sender reputation and may struggle with new or carefully crafted attacks. This paper presents a conceptual AI-Based Email Security System that analyzes email content and available message metadata to identify suspicious messages and support safer handling. The proposed approach combines text preprocessing, machine-learning classification, phishing indicators, URL and attachment checks, and a risk-based decision layer. Emails can be categorized as legitimate, spam, phishing, or requiring further review. The design also includes user feedback, explainable warnings, privacy-conscious processing, and periodic model evaluation. The system is intended as a defensive support tool rather than a guarantee that every malicious email will be detected. Its effectiveness would depend on representative training data, secure implementation, and continuous testing.
“AI-Based Code Generation and Self-Testing System”
Software development often requires programmers to write repetitive code, understand unfamiliar requirements, and spend considerable time identifying and correcting errors. Recent advances in artificial intelligence have enabled systems to generate source code from natural-language instructions. However, generated code may contain syntax errors, incorrect logic, insecure patterns, or incomplete handling of edge cases. This paper proposes an AI-Based Code Generation and Self-Testing System that accepts a programming task in natural language, generates a candidate solution, prepares or selects test cases, executes the code in a controlled environment, and analyses the results. The proposed workflow combines prompt interpretation, code generation, static validation, test-case generation, sandboxed execution, and feedback-based refinement. The system is intended to support developers and learners by reducing repetitive effort and providing an initial validation cycle. It does not replace code review or guarantee correctness; execution safety, test coverage, model limitations, and human verification remain important. This paper presents a proposed methodology and evaluation plan. Quantitative performance claims should be added only after implementation and testing.
AI Based Phishing Website Detection Using Machine Learning
Phishing websites are a common cybersecurity threat that trick users by pretending to be legitimate websites and attempting to obtain sensitive information. Traditional methods, such as blocklists and rule-based techniques, may not always detect newly created or modified phishing websites. Machine learning provides an effective approach to phishing detection by learning patterns from examples of legitimate and phishing websites. This paper presents a conceptual framework for AI-based phishing website detection using machine learning. The framework includes collecting phishing and legitimate website data, extracting important URL and website features, preprocessing the data, selecting relevant features, applying machine learning classification, and evaluating the detection performance. Recent research on feature engineering, feature selection, machine learning, deep learning, XGBoost-based detection, and phishing campaign detection is also reviewed. The proposed framework aims to provide a simple and structured approach for distinguishing phishing websites from legitimate websites while considering the changing nature of phishing attacks. Since this study is research and visualization oriented, the proposed system is presented as a conceptual framework rather than a developed application or trained software model.
AI Based Password Strength Analyzer
Passwords remain one of the most common methods of user authentication, but weak and reused passwords continue to expose accounts to guessing, credential stuffing, and brute-force attacks. This paper proposes an AI Based Password Strength Analyzer that evaluates password quality and provides understandable recommendations for improvement. The system combines conventional password-strength indicators—such as length, character variety, predictable patterns, and common-password matches—with a machine-learning component that can learn patterns associated with weak passwords from appropriately prepared datasets. Rather than treating a password as simply valid or invalid, the analyzer assigns a strength category and explains which characteristics influenced the assessment. Privacy is a central design requirement: password text should be processed locally whenever possible, should not be stored or logged, and should never be transmitted to a server without a clear security justification. The paper describes the proposed architecture, analysis workflow, implementation approach, evaluation plan, and limitations. The system is intended as a decision-support tool that encourages safer password choices; it does not replace multi-factor authentication, password managers, secure storage, or established authentication controls.
MALWARE DETECTION USING MACHINE LEARNING
Malware is a major cybersecurity threat because malicious files and programs can steal information, damage systems, disrupt services, and provide unauthorized access. Traditional signature-based detection methods are useful for known threats but may struggle to identify previously unseen or modified malware. This paper presents a machine-learning-based approach for classifying files or executable samples as benign or malicious using extracted features. The proposed workflow includes dataset collection, data cleaning, feature preparation, model training, testing, and performance evaluation. Candidate classifiers may include Decision Tree, Random Forest, Support Vector Machine, or other suitable models; the final model should be selected using measured results on a held-out test set. Evaluation should report metrics such as precision, recall, F1-score, accuracy, and confusion matrix. The approach is intended to support automated malware screening, while recognizing that model performance depends on dataset quality, feature selection, and the ability of malware to change over time. All experimental values in this draft are placeholders and must be replaced with results from the student's actual implementation.
official tracking system in AI/ML
Online businesses must answer product questions, explain policies, guide purchases and resolve order-related concerns across multiple channels. Human-only support can be costly and slow during peak demand, while rigid rule-based chatbots struggle with varied wording and multi-step questions. This paper presents a conceptual design for an Artificial Intelligence (AI) Chat Assistant for Online Business that combines natural language understanding, retrieval from approved business knowledge, controlled access to commerce services and escalation to human staff. The proposed assistant interprets customer intent, retrieves relevant product and policy information, generates grounded responses, and invokes permitted actions such as checking order status only after appropriate authentication. The architecture emphasizes factual consistency, privacy, security, accessibility and measurable service quality. It also describes a development methodology, evaluation metrics, risk controls and a staged deployment plan. Since this is a design-oriented study, no original performance results are claimed; the proposed evaluation must be conducted using representative customer queries and verified business data.
Databse Query Optimization
Large-scale applications generate high volumes of database requests and manage datasets that grow continuously. Inefficient queries increase latency, CPU utilization, memory pressure, storage I/O, and infrastructure cost. Query optimization selects and shapes database operations so that results are produced efficiently without changing their meaning. This paper reviews index design, query rewriting, join strategies, statistics, pagination, connection pooling, caching, partitioning, and workload monitoring. It emphasizes measurement-driven optimization: establish a baseline, inspect execution plans, change one factor at a time, and evaluate read and write effects under representative workloads.
Fraud transaction detection
Digital payments, online shopping, mobile banking, and card-not-present transactions have increased the speed and convenience of financial services. These channels also create opportunities for fraudulent activity, including stolen-card use, account takeover, identity misuse, and suspicious transaction patterns. This paper presents a proposed fraud transaction detection framework that combines transaction-data preparation, rule-based checks, machine-learning risk scoring, and human review. Transaction attributes such as amount, time, merchant category, location consistency, and recent activity can be transformed into features for a classification model. The system assigns a risk score and routes high-risk transactions for additional verification instead of treating every unusual transaction as confirmed fraud. The paper outlines the architecture, workflow, candidate algorithms, evaluation measures, privacy safeguards, and deployment considerations. Since no institution-specific dataset or completed experiment was supplied, performance values are not claimed; the model must be evaluated on representative, appropriately protected data before operational use.
Ai powered crop disease detection system
Crop diseases can reduce agricultural productivity and cause financial losses when symptoms are not recognised early. Farmers may have limited access to timely expert diagnosis, while visual symptoms can be difficult to distinguish across crops and growing conditions. This paper proposes an AI Powered Crop Disease Detection System that analyses images of crop leaves and provides a preliminary indication of possible diseases. The proposed workflow includes image acquisition, preprocessing, feature learning using a convolutional neural network (CNN) or a suitable transfer-learning model, disease classification, and a user-facing result with basic preventive guidance. The system is intended to support farmers and agricultural advisers rather than replace laboratory testing or expert judgement. A prototype may use a curated, labelled image dataset and a web or mobile interface for image submission. Evaluation should examine classification precision, recall, F1-score, robustness to lighting and background variation, and response time. No experimental performance results are claimed because the proposed system has not yet been benchmarked. The design emphasises reliable data, clear limitations, and responsible use of recommendations.
Agentic AI: The AI That Talks and Acts
Most AI tools only talk: they answer a question and stop. Agentic AI systems go further, because they plan steps, call tools and act on other software. This is useful, but a wrong answer from a chatbot is only text, while a wrong action from an agent can change a database or move money. This paper reviews how agentic systems are built, where they are used and how well they perform in published tests. We reviewed 29 sources (research papers, open standards and official rules) under six themes: reasoning, tool use, memory, protocols, safety and business use. Acting helps—ReAct beat earlier methods by 34% and 10% on two benchmarks—yet success on realistic tasks is still low: the best web agent completed 14.41% of tasks against 78.24% for humans. From the gaps found we propose a five-layer framework with a human-approval gateway and check it analytically against published cases. No software was built, and all numbers are quoted from the original authors. We conclude that agents work best when humans keep control of important decisions.
Re-imagining the Future of Work through Human-AI Collaboration
By combining the artificial intelligence (AI) and human experts knowledge is changing the way of how work is designed and executed. This is different from traditional automation techniques that only handle routine , regular tasks as Human-AI Collaboration integrates the speed and analysis capability of machines with human decision-making, empathy, and ethical oversight. This paper looks at recent studies on HAIC. It finds three major problems: Future-oriented thinking , differences between sectors and how to manage things together. It also propose a five-level system for arranging tasks, calculating trust, keeping human control, and checking ethics. The framework is studied using three case studies about customer support, management consulting, software development. It looks at tradeoffs, effects on workers, and rules. The evidence shows that when tasks are easy for AI, can be changed beyond what AI can do, and always need human checks, there are big benefits. The paper says that the future of work will involve teamwork between humans and AI, working together in a cycle with feedback. This framework is just an idea and has not been tried in a real company.
Phishing website detection using machine learning
Phishing websites imitate legitimate online services to trick users into revealing passwords, payment details, and other sensitive information. Static blacklists are useful for known threats but may not recognize newly created or rapidly modified sites. This paper presents a proposed machine-learning framework that classifies a website URL as likely legitimate or potentially phishing by examining lexical, domain, and— where safely available—HTML-based features. The workflow includes data validation, leakage-aware preprocessing, model training, threshold selection, and user-facing risk explanations. Random Forest, Support Vector Machine, and gradient-boosting models are considered as candidate classifiers. The evaluation plan uses precision, recall, F1-score, ROC-AUC, PR-AUC, false-positive rate, and inference time, with special attention to class imbalance and temporal testing. This paper is a design proposal; it does not claim newly conducted experiments or measured performance. The intended outcome is an understandable prototype that supports safer browsing while preserving human judgment and privacy.
Biomarker Discovery and Clinical Trial Optimization Driven By Artificial Intelligence
Biomarkers are measurable biological characteristics that can support disease diagnosis, prognosis, treatment selection, and monitoring of therapeutic response. Their discovery increasingly depends on high-dimensional and heterogeneous data such as genomics, transcriptomics, proteomics, medical imaging, electronic health records, and wearable-device measurements. Artificial intelligence (AI) can help identify patterns across these sources and support more efficient clinical-trial planning, including patient stratification, eligibility screening, recruitment forecasting, endpoint assessment, and risk monitoring. This paper presents a conceptual framework connecting AI-enabled biomarker discovery with clinical-trial optimization. The framework covers data governance, preprocessing, multimodal feature integration, candidate selection, validation, trial feasibility, and continuous monitoring. It emphasizes that computational performance alone does not establish clinical usefulness: candidate biomarkers require independent replication, analytical validation, clinical validation, and evidence of utility in a clearly defined context of use. The paper synthesizes published literature and regulatory principles; it does not report a newly conducted clinical study or claim original performance results. Responsible implementation requires privacy protection, representative data, transparent documentation, clinician oversight, and prospective evaluation before AI outputs influence patient care.
HEART DISEASE PREDICTION USING MACHINE LEARNING
Phishing is a social-engineering threat in which attackers impersonate trusted organisations or individuals to persuade users to disclose information, open harmful attachments, or visit deceptive websites. Conventional filters can be challenged by newly created domains, changing message patterns, and carefully crafted text. This paper proposes an AI-Powered Phishing Detection system that analyses selected email, URL, and message features to estimate whether an item may be suspicious. The workflow includes input validation, feature extraction, machine-learning classification, risk scoring, explanation of influential indicators, and a user-facing warning or review action. Candidate features include URL structure, domain signals, message wording, sender metadata, and selected email-header indicators where available. The system is a defensive screening aid, not a guarantee that a message is safe or malicious. The paper describes architecture, workflow, implementation, evaluation, applications, and limitations. Diagrams and chart values are illustrative, not experimental results.
Smart Water Quality Monitoring System with AI
Water quality is an important factor for aquatic ecosystems, agriculture, aquaculture and human activities. Dissolved oxygen (DO) is one of the important indicators of water condition because changes in its concentration can affect aquatic organisms and indicate changes in the surrounding environment. The selected Base Paper 5, “Measuring dissolved oxygen in pond water with high-resolution by optical gas sensor,” presents a high-resolution optical DO sensor based on a nanoporous anodic aluminum oxide (AAO) membrane and Platinum(II) meso-tetrakis (pentafluorophenyl) porphyrin (PtTFPP). The reported response of the sensor was 26.2 on the AAO membrane compared with 6.5 on a glass substrate, with a response time of approximately 16 s and recovery time of approximately 190 s. This seminar research proposes an extension of the DO monitoring concept into a Smart Water Quality Monitoring System with A
Adaptive phishing URL Detection using feature stability and uncertainty based Decision making Information technology
Phishing URLs are designed to mislead users into visiting deceptive websites and disclosing sensitive information. Blacklist-based systems can identify previously reported addresses, but newly created or modified URLs may not yet have a reputation record. Machine-learning classifiers can identify patterns in URL strings; however, their performance may depend on dataset composition, feature selection, and changes in attacker behaviour. This paper proposes an adaptive phishing URL detection pipeline that combines lexical and structural feature extraction, feature-stability analysis, supervised classification, and uncertainty-based decision making. Feature stability is assessed across repeated or time-separated training samples to identify features that remain consistently informative. The classifier produces a Phishing or Legitimate label only when its confidence satisfies a validation-selected threshold; otherwise, the system returns Uncertain for further review. The paper defines an evaluation protocol using precision, recall, F1-score, ROC-AUC, calibration, coverage, and covered error rate, with domain- and time-aware splits where feasible. This work presents a proposed methodology and research design. The model has not yet been implemented or experimentally evaluated, and no quantitative performance improvement is claimed.
DALS-IDS: Drift-Aware Lightweight Stacking Intrusion Detection for Smart Transport Systems
Smart transport systems combine in-vehicle Controller Area Network (CAN) communications, roadside infrastructure, and connected network services. These components can be exposed to changing traffic patterns and intrusion attempts, making a detector trained on a fixed dataset potentially unreliable over time. This paper proposes DALS-IDS (Drift-Aware Lightweight Stacking Intrusion Detection System), a research framework for intrusion detection across smart-transport and supporting network environments. DALS-IDS combines LightGBM, ExtraTrees, and Random Forest base learners with feature subspacing and a Logistic Regression meta-learner trained using out-of-fold predictions. To address changing data distributions, a sliding-window performance monitor updates ensemble weights, while ADWIN is used as a concept-drift detector to trigger weight recalculation or recent-window meta-learner retraining. A SHAP-based explanation layer reports influential features for each alert. The design emphasizes leakage-free preprocessing, CPU-only efficiency measurement, per-class evaluation, temporal testing, and cross-dataset generalization. Planned evaluation includes CAN-bus datasets and network intrusion datasets, subject to verification of dataset availability, schema compatibility, and permissible feature mapping. This paper presents a proposal and planned evaluation methodology; it does not claim experimental results or validated performance improvements.
MACHINE LEARNING IN EDUCATION: INNOVATIONS, IMPACTS,AND ETHICAL CONSIDERATIONS
Machine Learning (ML) has recently emerged as a powerful tool with significant potential to revolutionize education, bringing about fundamental changes in pedagogy and research. Its applications span various domains within academia, including administration, instructional method enhancement, and grade prediction. This study illustrates how ML can enhance the effectiveness of research, instruction, and study strategies by adapting to student needs and leveraging new communication tools within virtual learning environments. The literature review encompasses various empirical studies outlining the integration of Artificial Intelligence (AI) and ML within the educational sector. This research paper maps out how educators are leveraging these capabilities while documenting the associated risks and learning outcomes across various educational tiers. You can also refer to the original context in the file named 'Suhani Research Paper OG.pdf' for a comprehensive view. We advocate that current educational practices in engineering and computing with GenAI can indicate to us a roadmap of its potentialities, uses, and risks from the standpoint of both teachers and students. Our purpose with the outcomes and conclusions of this scoping review is to support educators, faculty members, and other stakeholders in engineering and computing education to co-create educational methodologies that articulate GenAI with curricula, AI literacy, and prompt engineering encompassing students’ learning domains such as cognitive, affective, or behavioral. This sample paper is mapped against the IJREK standard student template format and expands on the empirical findings.
Ecommerce Product Recommendation
E-commerce platforms present customers with large and frequently changing product catalogues. Finding relevant products can require substantial browsing, and generic rankings may not reflect an individual shopper’s interests, current needs, or budget. This paper proposes an e-commerce product recommendation system that combines product-content similarity with behavioural signals to create personalized, context-aware product suggestions. The design includes catalogue preparation, interaction-event processing, candidate generation, ranking, availability filtering, and feedback collection. A hybrid approach is proposed because content-based recommendations can serve new or lesspopular products, while collaborative signals can identify patterns across users with similar preferences. The system also includes practical safeguards such as privacy-aware data handling, diversity controls, explanations, and monitoring for popularity bias. A prototype could be developed using a web interface, an API service, a product database, and a recommendation module. Evaluation is planned using ranking metrics and usability measures; no experimental performance is claimed in this conceptual paper. The proposed design aims to improve product discovery while keeping recommendations understandable and under customer control.
Student Learning Using Ai
Artificial intelligence (AI) is increasingly being explored as a way to make digital learning more responsive to students’ individual needs. This paper proposes an AI-assisted student learning system that combines learner preferences, topic-level progress, practice responses, and teacher feedback to recommend suitable learning resources and next-step activities. The proposed framework includes a learner interface, a content repository, a recommendation and feedback engine, progress analytics, and teacher oversight. Rather than treating a single score as a complete picture of ability, the system uses repeated learning evidence to identify topics that may need revision and to suggest practice at an appropriate level. The design emphasizes explainable recommendations, privacy, accessibility, and human supervision. A prototype could be developed using a web interface, a structured content store, and rulebased or machine-learning components selected according to available data. Evaluation should measure recommendation relevance, consistency, usability, learner engagement, and learning outcomes through a controlled and ethically approved study. No experimental performance is claimed in this conceptual paper; the framework and evaluation plan are presented for future implementation.
Ai-Powered Smart Agri
Agriculture is influenced by changing weather, soil conditions, crop selection, plant disease, and market variability. Farmers may need to consult multiple sources to obtain timely and understandable guidance. This paper presents a proposed AIpowered smart agriculture framework that combines farmer-provided information with weather services, crop and soil knowledge, plant-health image analysis, and market information. A web or mobile interface collects a location, crop details, and the farmer’s concern. An application layer validates the inputs and coordinates the available services. The recommendation module then presents context-sensitive suggestions with their data sources, confidence indicators where supported, and appropriate cautions. The design emphasizes explainability, accessibility, privacy, and the involvement of local agricultural experts for consequential decisions. A prototype may be implemented using a REST API, a structured knowledge base, and separately deployable analytical services. Evaluation should measure functional correctness, response time, recommendation relevance, usability, and robustness under missing or outdated data. Because this paper proposes a framework rather than reporting completed field trials, no experimental accuracy or yield improvement is claimed.
Fake Profile Detection on Social media
Social media platforms enable communication, news sharing, and online communities, but fake or inauthentic accounts may be used for spam, impersonation, scams, coordinated manipulation, and misleading content. This paper proposes a machine-learning-based Fake Profile Detection system that evaluates available account and activity features and flags profiles that may require further review. The planned workflow includes dataset selection, data validation, preprocessing, feature engineering, model training, threshold selection, and evaluation. Candidate methods include Logistic Regression, Decision Tree, Random Forest, and anomaly-detection techniques when suitable. Because the prevalence of fake accounts and the reliability of labels can vary by dataset, accuracy alone is insufficient; precision, recall, F1-score, confusion matrices, and precision-recall analysis should be considered. This is a proposed academic prototype: no measured accuracy or operational effectiveness is claimed. Any real deployment should protect user privacy, limit data collection, reduce bias, and provide human review before consequential action.
GENERATIVE AI IN ENGINEERING AND COMPUTING EDUCATION: A SCOPING REVIEW OF EMPIRICAL STUDIES AND EDUCATIONAL PRACTICES
Since the release of diverse generative AI (GenAI) tools such as ChatGPT, Google Gemini, DALL-E, and GitHub Copilot, there has been much debate around the impacts and implications of these tools on education. Currently, extant literature remarks on the affordances, challenges, and opportunities of GenAI, but few studies report and analyze empirical studies and educational practices coming up by GenAI usage in learning settings. Then, in this Scoping Review (ScR) based on 146 studies retrieved from the databases SCOPUS, Web of Science (WoS), and ERIC, we analyzed the implications of integrating GenAI in engineering and computing education from K-12 to tertiary levels. We adopted an approach starting from the bibliometric features of the studies in terms of authors, cites, years, or cluster topics, and navigating to the identification of methodologies, strategies, AI literacy instruments and guidelines, learning outcomes, and students’ and teachers’ perceptions, among other features. We advocate that current educational practices in engineering and computing with GenAI can indicate to us a roadmap of its potentialities, uses, and risks from the standpoint of both teachers and students, and this could help us to create more reflexive methodologies that enhance the teaching-learning process based on the evidence. Our purpose with the outcomes and conclusions of this scoping review is to support educators, faculty members, and other stakeholders in engineering and computing education to co-create educational methodologies that articulate GenAI with curricula, AI literacy, and prompt engineering encompassing students’ learning domains such as cognitive, affective, or behavioral. This sample paper is mapped against the IJREK standard student template format.
Smart Transport System
Urban and semi-urban transport systems face challenges such as traffic congestion, delays, inefficient route planning, limited real-time information, and difficulty coordinating transport services. This paper proposes a Smart Transport System that uses digital technologies to improve access to transport information and support more efficient travel planning. The proposed system may combine route information, vehicle or service updates, location-aware features, and a user interface that presents useful travel details. Depending on available data sources, it can help users view routes, estimated arrival information, and service notifications. The system is intended as a decision-support application and its reliability depends on data freshness, connectivity, and the accuracy of the underlying transport information. A prototype may be evaluated for response time, route-information accuracy, usability, and availability. No experimental performance results are claimed because the proposed system has not yet been benchmarked. The design emphasises data protection, accessibility, and clear communication when real-time information is unavailable.
Personalized Healthcare Recommendation System
People have different health needs, daily routines, medical histories, and lifestyle choices. As a result, general health advice may not be equally useful to everyone. This paper presents a proposed Personalized Healthcare Recommendation System that uses information provided by users to offer relevant wellness suggestions. The system considers factors such as age, gender, reported symptoms, diet, physical activity, lifestyle, and medical history. After processing these details, it can present suggestions related to healthy habits, nutrition, physical activity, and preventive care. The main purpose is to make health information easier to understand and more relevant to individual users. The system is intended to support health awareness and does not replace professional medical diagnosis or treatment.
Explainable AI for Detecting and Explaining Software Bugs in Java Applications
Software bugs can cause incorrect results, crashes, security weaknesses, and expensive maintenance work. Traditional testing and static-analysis tools can identify many defects, but their warnings may be difficult to prioritise or understand, especially in large Java applications. Artificial intelligence (AI) can help rank suspicious code and recognise patterns associated with defects; however, a prediction is less useful when developers cannot understand why it was produced. This paper proposes an explainable AI (XAI) workflow for detecting and explaining potential bugs in Java applications. The proposed approach combines source-code features, static-analysis findings, test and version-history signals, and a prediction component with explanations tied to observable evidence such as a risky code path, a missing null check, a resource-handling pattern, or a related test failure. The design aims to present a confidence score, relevant code location, supporting signals, and suggested next steps while keeping the developer responsible for confirming the defect. The paper discusses the problem, related approaches, architecture, methodology, evaluation plan, applications, limitations, and future work. Since this is a proposed design, performance improvements must be established through experiments rather than assumed.
Emotion detection system
Emotion detection is an area of affective computing that attempts to estimate emotional cues from observable signals such as facial expressions, voice characteristics, or text. This paper proposes an academic Emotion Detection System that accepts a selected input modality, processes it, extracts relevant features, and uses a machine-learning or deep-learning model to estimate an emotion category. The design can support applications such as human-computer interaction, learning interfaces, and user research, provided that users understand the system's limitations. The proposed workflow covers input capture, preprocessing, feature extraction, classification, result presentation, and basic evaluation. The system should communicate that its output is an estimate rather than a definitive statement about a person's internal emotional state. Privacy, informed consent, bias, accessibility, and secure handling of audio, image, or text data are considered throughout the design. Architecture and workflow diagrams, a conceptual output chart, and an implementation and evaluation plan are provided. The figures are illustrative; no experimental accuracy is claimed.
AI based email security system
Email is a central channel for personal and organizational communication, but it is also exploited to distribute spam, phishing links, fraudulent requests, and malicious attachments. This paper proposes an AI-based email security system that analyzes message text, sender-related indicators, URLs, and attachment metadata to estimate whether an email is legitimate, spam, phishing, or potentially malicious. The proposed workflow combines preprocessing, feature extraction, machine-learning classification, risk scoring, and explanations for flagged messages. A layered design is recommended because no single indicator is reliable in every case: language patterns, sender authentication signals, suspicious URL characteristics, and attachment properties should be considered together. The system should issue a warning or quarantine recommendation rather than automatically treating every flagged message as malicious. Evaluation can measure precision, recall, F1-score, false-positive rate, detection latency, and robustness to changing attack patterns. As no implemented prototype or measured dataset results were provided, this paper presents a proposed design and evaluation plan rather than claiming numerical performance.
DIABETES / HEART DISEASE PREDICTION USING MACHINE LEARNING
Diabetes and heart disease are important health conditions for which assessment of risk factors may support timely discussion with healthcare professionals. This paper proposes a machine-learning-based prediction system that uses selected attributes from an appropriate, documented dataset to classify records into dataset-defined categories for diabetes or heart disease. Possible algorithms include Logistic Regression, Decision Tree, Random Forest, and Support Vector Machine. The workflow includes data cleaning, preprocessing, model training, testing, and presentation through a simple interface. The system is intended for educational and research purposes only. Its output is not a diagnosis, does not replace clinical assessment, and must not be used to make treatment decisions. This is a proposed design and evaluation plan; no measured accuracy or clinical effectiveness is claimed.
CFTR Carrier Screening
Cystic fibrosis transmembrane conductance regulator (CFTR) carrier screening is a genetic testing approach used to identify individuals who carry variants associated with cystic fibrosis, an inherited disorder that affects multiple organs, particularly the respiratory and digestive systems. Carrier screening is relevant to reproductive planning because two carriers of disease-causing CFTR variants may have a risk of having an affected child. This paper reviews the biological basis of CFTR-related disease, the role of carrier screening, commonly used testing approaches, interpretation of results, and ethical considerations. It also proposes an information-management workflow suitable for an educational technology project: consent and intake, test-order tracking, secure result handling, interpretation support, and referral to qualified genetics professionals. The workflow is not a diagnostic system and cannot replace laboratory testing or genetic counselling. Screening performance depends on the variants covered, the test method, ancestry-related variant frequencies, and the clinical context. The paper emphasizes informed consent, privacy, residual risk after a negative result, and the need for professional interpretation.
Ai based phishing website detection
Phishing websites imitate legitimate websites to deceive users into revealing sensitive information such as login credentials, banking details, and personal data. Traditional detection methods often depend on manually maintained blacklists, which may not identify newly created or rapidly changing malicious websites. This paper presents a proposed AI-based phishing website detection system that analyzes URL characteristics and selected website signals to classify a submitted website as potentially legitimate or suspicious. The system extracts features such as URL length, use of an IP address, unusual symbols, domain-related indicators, HTTPS presence, and selected page-level signals where safely available. These features can be processed by a supervised machine-learning classifier and combined with transparent rule-based checks. The output is presented as a risk label with a brief explanation, while uncertain cases are treated cautiously. The proposed design aims to support safer browsing and provide a modular foundation for further evaluation. The document describes a research-paper sample; actual dataset details, model selection, experimental results, and references must be verified against the author's implementation before publication.
STUDENT PERFORMANCE PREDICTION
Student performance prediction applies educational data mining and machine learning to estimate academic outcomes from information such as attendance, assessment marks, assignment completion, prior grades, and participation. The purpose of this paper is to describe a practical framework that can help educators identify students who may benefit from additional academic support. The proposed workflow covers data collection, preprocessing, feature engineering, model training, evaluation, and presentation of predictions through an interpretable dashboard. Candidate methods include Logistic Regression, Decision Tree, and Random Forest models. Performance should be assessed with metrics suited to the task, including accuracy, precision, recall, F1-score, and error analysis where the target is a numerical mark. The paper also considers data privacy, bias, explainability, and the risk of treating a prediction as a fixed judgment about a learner. No original experimental dataset or measured model result is claimed in this conceptual study. Instead, it presents a design that can be implemented and evaluated using institutionally approved data, with predictions used as decision support rather than as a replacement for teacher judgment.
expense manager
Managing daily expenses is an important part of personal financial planning. When spending is recorded manually in notebooks or scattered applications, users may find it difficult to understand where money is spent, follow a monthly budget, and identify recurring financial commitments. This paper proposes an Expense Manager system that enables users to record income and expenses, classify transactions, define budgets, and review summaries through a simple digital interface. The system is designed around a transaction module, category management, budget monitoring, reports, and secure user access. Monthly summaries and visual charts can help users compare planned budgets with actual spending, while reminders can support regular tracking of recurring payments. The proposed design emphasizes usability, data validation, privacy, and reliable storage. A staged evaluation plan is outlined using functional tests, usability feedback, calculation accuracy, and system reliability. This paper presents a design and implementation proposal; it does not claim measured user-study outcomes or financial savings without experimental evidence.
FAKE PROFILE DETECTION ON SOCIAL MEDIA USING MACHINE LEARNING
Social media supports communication, learning, entertainment, and business, but fake or misleading accounts may spread spam, impersonate people, manipulate engagement, or direct users to fraudulent content. Manual identification is difficult at scale. This paper proposes a supervised machine-learning workflow to classify profiles as potentially genuine or potentially fake using available profile features. The planned pipeline includes a labelled dataset, data cleaning, feature preparation, train-test splitting, and a Random Forest classifier. Candidate features include profile-picture availability, biography and username characteristics, follower/following counts, post count, and account age when lawfully available. Performance will be assessed with precision, recall, F1-score, accuracy, and a confusion matrix. False accusations and missed fake accounts are both important risks. This is a research plan, not a completed experiment; no project-specific performance score is claimed.
DATA CONSISTENCY IN DISTRIBUTED DATABASE SYSTEMS
Distributed database systems store and process data across multiple nodes to improve availability, scalability, and geographic reach. However, maintaining a consistent view of data across replicas is challenging when network delays, concurrent updates, node failures, and partitions occur. This paper reviews the foundations of data consistency in distributed databases and presents a conceptual framework for selecting and managing consistency guarantees according to application needs. It discusses strong consistency, eventual consistency, causal consistency, and session-level guarantees, together with replication strategies, quorum-based operations, conflict resolution, version tracking, and monitoring. A proposed architecture separates request coordination, replication, policy enforcement, conflict handling, commit history, and replica-health monitoring. A workflow diagram illustrates how a write request can be routed, replicated, acknowledged, and later read under a defined policy. The paper also outlines use cases, trade-offs, implementation considerations, and an evaluation plan covering read freshness, write latency, conflict rate, availability, and recovery time. The discussion emphasizes that no single consistency model is optimal for every workload: financial records and inventory may require stronger guarantees, while feeds and analytics may tolerate bounded staleness. This paper is a conceptual study and does not claim experimental measurements from a deployed database.
AI In Healthcare
Artificial Intelligence (AI) is transforming healthcare by supporting clinical decision-making, medical image analysis, disease prediction, personalized treatment, drug discovery, and remote patient monitoring. Machine learning, deep learning, natural language processing, and computer vision enable healthcare systems to extract useful information from medical images, electronic health records, laboratory reports, and physiological signals. These capabilities can assist healthcare professionals in identifying potential abnormalities, estimating clinical risks, and improving the organization of patient care. However, the practical adoption of AI introduces important challenges, including data privacy, algorithmic bias, limited interpretability, cybersecurity threats, regulatory compliance, and the need for clinical validation. An AI system that performs well on a development dataset may not achieve the same performance in a different hospital or patient population. Therefore, accuracy alone is insufficient to establish clinical usefulness. This paper reviews the major AI technologies used in healthcare, presents a generalized architecture for AI-enabled clinical systems, examines representative applications, and discusses their benefits, limitations, and ethical implications. It also explores future directions involving explainable AI, federated learning, multimodal models, and human–AI collaboration. The review emphasizes that AI should complement clinical expertise rather than replace professional judgment. Responsible implementation requires reliable data, independent validation, transparent evaluation, appropriate governance, and continuous monitoring after deployment. Index Terms—Artificial Intelligence, Healthcare, Machine Learning, Deep Learning, Medical Imaging, Clinical Decision Support, Explainable AI, Healthcare Data Privacy.
Protecting Digital Systems, Networks and Information: A Review of Cybersecurity Threats and Defensive Practices
Digital services support communication, education, finance, healthcare and business, making the confidentiality, integrity and availability of information essential. At the same time, phishing, malware, ransomware, credential attacks and weaknesses in network configuration expose users and organizations to disruption and data loss. This review examines common cybersecurity threats and summarizes practical defensive controls, including multifactor authentication, encryption, firewalls, endpoint protection, network segmentation, patch management, backups and security awareness. The review uses authoritative guidance from the National Institute of Standards and Technology (NIST), the Indian Computer Emergency Response Team (CERT-In), and technical reporting on the WannaCry ransomware outbreak. The analysis indicates that no single product can provide complete protection; risk reduction depends on layered controls, clear governance, timely maintenance and tested incident-response processes. The paper concludes with a practical, risk-based security approach suitable for educational institutions and small organizations. It is a literature-based review and does not claim original experimental measurements.
Smart E-Learning Platform Using Artificial Intelligence
Online learning platforms provide access to educational content, but many systems deliver the same material to every learner and offer limited support for identifying individual learning needs. This paper proposes a Smart E-Learning Platform using Artificial Intelligence (AI) that aims to personalize learning through learner profiles, progress tracking, recommendation methods, automated practice feedback, and a dashboard for learners and instructors. The proposed workflow collects learning activity and assessment results, processes relevant learning indicators, and recommends suitable resources or practice activities. Possible AI methods include rule-based recommendation as a baseline, content-based filtering, and supervised models when sufficient labelled data is available. The platform is designed as an academic prototype and should be evaluated using measurable indicators such as recommendation relevance, task completion, learner satisfaction, and usability. No implementation results or improvement percentages are claimed in this proposal. Privacy, transparency, accessibility, and instructor oversight are important requirements. The system is intended to support teaching and learning, not to replace teachers or make high-stakes decisions about students.
AI-POWERED PERSONAL DATA ANALYST
Individuals create personal data through expenses, calendars, fitness applications, learning platforms, digital notes, and other everyday services. Although these records can support better decisions, they are often scattered across applications and presented without a unified explanation. This paper proposes an AI-powered personal data analyst that consolidates user-authorized data, organizes it into a consistent structure, and produces understandable summaries, trends, and recommendations. The proposed framework includes data ingestion, validation, normalization, privacy controls, analytical processing, optional machine-learning components, and an interactive dashboard. Descriptive statistics and rule-based insights form the initial baseline; anomaly detection and forecasting can be introduced only when sufficient, suitable data are available. The design emphasizes transparent explanations, user consent, data minimization, and a clear distinction between observed facts and model-generated estimates. The paper outlines the system architecture, workflow, implementation options, evaluation metrics, limitations, and future scope. It is a design proposal and does not claim measured performance or completed user testing. A practical prototype should be evaluated with consented data, representative test cases, and privacy and usability checks before wider deployment.
smart parking management system
Finding an available parking space can be difficult in busy campuses, commercial areas, hospitals, and urban centres. Conventional parking management often depends on manual observation or static information, which may be outdated and can lead to unnecessary circulation, wasted time, and local congestion. This paper proposes a Smart Parking Management System that monitors parking-slot occupancy and communicates availability to drivers and parking administrators. The conceptual design combines slot-level sensors or camera-based detection, a controller or edge device, a network connection, a database, and a web or mobile interface. The system can display occupied and vacant spaces, support slot allocation or reservation, record entry and exit events, and provide administrative reports. A modular architecture is described along with a methodology for data collection, occupancy detection, availability updates, and evaluation. Relevant evaluation measures include occupancy-detection precision and recall, status-update latency, booking conflicts, and usability. Security, privacy, sensor faults, connectivity failures, and accessibility are considered as design requirements. This manuscript presents a proposed academic system and evaluation plan; it does not claim experimental results or measured performance because a prototype evaluation has not been conducted for this paper.
DESIGN AND DEVELOPMENT OF A CAREER GUIDANCE SYSTEM FOR STUDENTS AFTER CLASS 10 USING A RULE-BASED RECOMMENDATION APPROACH
Selecting an educational pathway after Class 10 is an important decision for students. Options may include higher-secondary streams, diploma programmes, and vocational courses, each with different learning experiences, entry requirements, and future opportunities. This paper proposes the design and development of a rule-based career guidance system that helps students explore educational pathways through a transparent, preference-based process. The system collects information about preferred subjects, interests, learning activities, and goals. A knowledge base stores pathway descriptions and verified eligibility information, while an IF–THEN recommendation engine matches student responses to relevant options. The proposed prototype uses Python and Streamlit to provide an accessible web interface and explain why each option is suggested. Evaluation is planned around recommendation consistency, test-case performance, explanation clarity, and usability. The system is intended as a decision-support tool and not as a substitute for professional counselling. Actual effectiveness will be established through implementation and evaluation rather than assumed in advance.
The Rise of Agentic AI in Software Engineering: Transforming the Software Development Lifecycle (SDLC)
Agentic Artificial Intelligence (AI) is an emerging approach in which AI systems can plan tasks, use tools, maintain context, and execute multi-step workflows with varying levels of autonomy. In software engineering, these capabilities may reshape the Software Development Lifecycle (SDLC), from requirements analysis and design to coding, testing, deployment, and maintenance. This paper examines how agentic AI can support development teams through task decomposition, repository-aware code generation, test creation, defect investigation, documentation, and operational feedback. It presents a conceptual architecture in which an orchestrator coordinates specialist agents and tools, while policy checks and human review govern consequential actions. The paper compares traditional automation, AI-assisted coding, and agentic workflows; identifies potential benefits and risks; and proposes an evaluation framework based on quality, productivity, reliability, security, and review effort. Agentic systems can reduce repetitive work and accelerate feedback, but they may also produce incorrect changes, expose sensitive data, or take unsafe actions when permissions and validation are weak. Accordingly, the proposed approach treats autonomy as bounded and reversible, uses tests and policy gates, and preserves human accountability. This paper is a conceptual design and literature-informed discussion; it does not claim experimental results from a deployed system.
A Smart Property Finder and In-Person Verification System for Transparent Real Estate Transactions
Real estate searches are often complicated by incomplete property information, duplicate advertisements, misleading photographs, uncertain ownership claims, and difficulty arranging physical visits. These issues can reduce trust between buyers, tenants, property owners, and intermediaries. This paper proposes a Smart Property Finder and In-Person Verification System designed to make property discovery more transparent by combining structured listings, search filters, owner or agent details, visit scheduling, and an auditable verification-status workflow. The proposed system allows users to compare properties using location, budget, property type, area, amenities, and availability. A verification module supports scheduled in-person inspection and records the outcome, observations, and status of each listing. The system is intended to distinguish information that is merely submitted by a user from information checked through a defined process. A modular web-based architecture can support user authentication, listing management, search, appointment booking, document review, feedback, and administrative oversight. The paper presents the system design, workflow, data model, security considerations, and an evaluation plan. Since the work describes a proposed system, performance values are not claimed as measured results. The approach aims to improve clarity, reduce avoidable visits, support informed decisions, and create a more accountable digital property-search experience.
SMART HOME ENERGY OPTIMIZATION USING AI
Residential electricity consumption varies with occupancy, weather, appliance use, and household routines. Conventional energy monitoring often provides limited visibility into which devices consume the most electricity and when demand is highest. This paper proposes an AI-assisted smart home energy optimization framework that combines smart-meter or appliance-level measurements, Internet of Things (IoT) communication, consumption analytics, and automated control recommendations. The system collects time-stamped energy data, validates readings, forecasts short-term demand, and identifies opportunities to shift flexible loads or reduce unnecessary consumption. A rule-based controller provides a transparent baseline, while machine-learning models may be evaluated when sufficient historical data are available. The design includes a dashboard for energy trends, estimated costs, alerts, and user-approved automation. Privacy, device security, reliability, and manual override are treated as core requirements. An evaluation plan is presented using prediction error, energy-use comparisons, peak-demand measures, and system reliability. This paper describes a proposed design and does not claim measured energy savings or field-tested model accuracy.
BLOCKCHAIN-BASED FRAMEWORK FOR SECURE AND TRANSPARENT DIGITAL DATA MANAGEMENT
Digital data is created, exchanged, and stored across organizations and online services. Conventional centralized systems can face challenges involving unauthorized modification, limited auditability, fragmented access control, and trust between independent parties. This paper proposes a blockchain-based framework for secure and transparent digital data management. The framework uses a permissioned distributed ledger to record data-related events, cryptographic hashes to support integrity verification, smart contracts to enforce agreed rules, and off-chain storage for large or sensitive files. Rather than storing confidential documents directly on a public ledger, the design keeps the underlying data in controlled storage and records only appropriate references, hashes, permissions, and audit events on-chain. The proposed architecture includes identity and access management, data submission, integrity checking, ledger recording, retrieval, and audit modules. Security, privacy, scalability, and governance are considered throughout the design. This paper provides a proposed architecture and evaluation plan; actual performance and security claims must be established through implementation and testing.
AI-Powered Resume Screening and Job Recommendation System
Recruitment teams may receive a large volume of resumes for a vacancy, making manual screening time-consuming and inconsistent. Job seekers may also struggle to identify vacancies that align with their skills, education, experience, and interests. This paper proposes an AI-powered system that extracts structured information from resumes, compares candidate profiles with job descriptions, and recommends relevant opportunities. The pipeline combines document parsing, text preprocessing, skill extraction, TF-IDF or sentence-embedding representations, and a transparent matching score. Candidate-facing results explain why a job was recommended, while recruiter-facing results support—rather than replace— human review. The design includes privacy controls, bias evaluation, and mechanisms for correcting extracted information. A reproducible evaluation plan uses extraction quality, Precision@K, Recall@K, ranking metrics, and subgroup error analysis. No experimental results are claimed because the proposed system has not been benchmarked for this manuscript.
AI-Based Threat Detection in Industrial OT Networks
Industrial Operational Technology (OT) networks connect programmable logic controllers, supervisory control and data acquisition (SCADA) systems, sensors, engineering workstations, and industrial control equipment. These systems support manufacturing, energy, water treatment, and other critical operations. Increasing connectivity between OT and information technology (IT) environments creates additional paths for cyber threats, while the operational consequences of disruption can be significant. This paper proposes an AI-based threat-detection framework designed for industrial OT networks. The proposed design combines passive collection of authorised network metadata and device logs, protocol-aware feature extraction, supervised machine-learning classification, anomaly detection, risk scoring, and human-reviewed alerts. The methodology prioritises OT-specific constraints such as predictable communication patterns, legacy devices, limited computing resources, availability requirements, and the risk of disrupting physical processes. The paper outlines a prototype architecture, candidate datasets and evaluation metrics, security and governance requirements, and possible extensions. No experimental results are claimed because the proposed system has not been implemented and benchmarked in a controlled industrial environment. The framework is intended as a defensive academic prototype and requires validation before operational deployment.
AI-Driven Phishing Email Detection: A Comparative Review of Ensemble Learning, Explainable AI, and Stylometric Analysis
Phishing email remains a persistent cybersecurity threat, and large language models (LLMs) now allow fluent phishing text to be produced at scale. This paper is a focused, literature-based comparative review of three recent studies that approach phishing email detection from different directions: a soft-voting ensemble with explainable-AI error analysis, a TF-IDF/SVM web platform with LIME explanations, and a stylometric classifier evaluated on GPT-4o-generated phishing emails alongside commercial spam filters. The studies are compared by objective, dataset, feature family, model, evaluation protocol, reported metrics, explainability and stated limitations. Because the studies use different corpora and protocols, their reported figures are presented with attribution and are not ranked. The comparison suggests that content/structural detectors and stylometric detectors have been evaluated on largely disjoint data, so whether either generalizes across conventional and AI-generated phishing is an open question raised by this review, not a verified research gap. The paper proposes a conceptual, unvalidated workflow that combines header/content/URL features, stylometric features, an ensemble classifier and LIME-based analyst review, and specifies a future experimental protocol with hypotheses, cross-dataset tests and ablations. No new experimental results are reported.
online course recommendation system
Online learning platforms provide a wide range of courses, but learners may find it difficult to select suitable courses from large catalogues. Generic course lists do not always reflect a learner’s goals, current skills, preferred learning style, available time, or prior learning. This paper proposes an Online Course Recommendation System that organises course information and recommends relevant learning options using learner preferences, course metadata, and interaction history where available. The proposed design combines content-based filtering with an optional collaborative-filtering component. Course descriptions and skill tags can be represented using keyword features or text embeddings, while learner profiles capture interests, experience level, preferred duration, and learning objectives. Recommendations should include understandable reasons, such as matching skills or selected topics, and should allow users to revise their preferences. The paper outlines the system architecture, data preparation, recommendation workflow, implementation plan, and evaluation metrics such as Precision@K, Recall@K, NDCG@K, catalogue coverage, and response time. Privacy, cold-start problems, popularity bias, accessibility, and outdated course information are considered. This manuscript describes a proposed academic system and evaluation plan; it does not report experimental results because no prototype evaluation has been conducted for this paper.
SmartBook: a hybrid filtering system for personalized book recommendation
Scholarly manuscript entitled 'SmartBook: a hybrid filtering system for personalized book recommendation'.
A PRIORITY-BASED PERSONALIZED STUDY SCHEDULING SYSTEM
Students often manage several subjects, assignments, examinations, and personal commitments at the same time. A fixed timetable may not reflect differences in task urgency, difficulty, learning goals, available time, or progress. This paper proposes a priority-based personalized study scheduling system that organizes academic tasks, ranks them using transparent criteria, and creates a practical study plan based on individual availability. The proposed system considers deadline proximity, task importance, estimated effort, topic difficulty, and learner progress. A scheduling module allocates time blocks, while a progress tracker updates the plan when tasks are completed or priorities change. The interface presents upcoming tasks, daily plans, reminders, and subject-wise progress. The design emphasizes explainable prioritization, user control, realistic workload limits, and privacy of student information. A rule-based scoring method is proposed as a baseline, with future scope for adaptive recommendations using historical completion patterns. An evaluation plan is described using schedule feasibility, task completion, ranking usefulness, and user feedback. This paper presents a design proposal and does not claim experimental improvements without a completed study.
Blockchain For Cybersecurity
Cybersecurity systems must protect digital identities, data, devices, and transactions against tampering, unauthorised access, and service disruption. Many conventional security architectures rely on centrally managed records and trusted administrators; these can become attractive targets and may provide limited visibility when logs are altered or distributed across organisations. This paper presents a proposed blockchain-supported cybersecurity framework that uses a replicated ledger, cryptographic hashes, digital signatures, and policy-based access to strengthen auditability and integrity. The design considers identity and access events, security-log integrity, configuration changes, and controlled sharing of threat indicators. Sensitive payloads are kept off-chain, while the ledger stores hashes, timestamps, references, and authorised approvals. The paper describes the motivation, architecture, workflow, implementation plan, evaluation metrics, applications, and limitations. Blockchain is treated as a supporting trust and audit mechanism rather than a replacement for encryption, secure software development, endpoint protection, or incident response. The proposal requires empirical testing before any claims about performance or security improvement can be made.
Multiagent AI Teaching Assistant for Personalization and Adaptive Programming Education
Programming education requires learners to develop conceptual understanding, problem-solving strategies, debugging skills, and the ability to transfer knowledge to unfamiliar tasks. A single teaching pathway may not meet the needs of learners who differ in prior experience, pace, confidence, and preferred forms of explanation. This paper proposes a Multiagent AI Teaching Assistant that personalizes programming education by coordinating specialized agents for tutoring, code analysis, assessment, feedback, and learner modelling. A coordinator combines learner goals and performance evidence to select the next activity, while a code-evaluation component examines submissions against tests and rubric criteria. The proposed system adapts explanations, hints, examples, and practice difficulty according to observed learner needs. A feedback loop updates a learner profile after each activity and supports gradual progression from guided examples to independent problem solving. The paper presents a conceptual 3D-style architecture diagram, a learning flowchart, functional modules, data entities, and a risk-aware evaluation plan. It emphasizes explainable recommendations, privacy, fairness, and human educator oversight. The work is a proposed design rather than a report of a completed classroom experiment; therefore, no measured performance improvements are claimed. Future implementation should compare learning outcomes, task completion, debugging ability, learner engagement, and teacher workload against a suitable baseline.
Machine Learning Model Design and Working for Hand-Drawn Diagram Recognition: Methods, Challenges, and Applications
Hand-drawn diagrams are a compact way to communicate processes, systems, and relationships, but converting them into machine-readable representations is difficult because symbols, text, and connectors vary substantially across people and drawing conditions. This paper presents a structured review and a conceptual machine-learning pipeline for hand-drawn diagram recognition (HDDR). The pipeline separates image preprocessing, symbol detection, text recognition, connector analysis, and graph reconstruction rather than treating a diagram as a single image-classification problem. We discuss traditional imageprocessing methods, convolutional neural networks, object detectors, segmentation networks, optical character recognition, and graph-based reasoning. We also describe evaluation measures and a reproducible experimental protocol that can be used to compare candidate models. The paper is a methods-oriented review and framework proposal; it does not claim new experimental results or measured accuracy. The analysis highlights error propagation between modules, limited and imbalanced datasets, ambiguous connector topology, and the need to evaluate semantic structure in addition to symbol classification. Potential applications include education, software design, engineering documentation, and digitisation of whiteboard workflows. Index Terms—hand-drawn diagram recognition, machine learning, computer vision, deep learning, optical character recognition, graph reconstruction, sketch understanding
Fraud Transaction Detection
Digital payment systems and online financial services enable fast and convenient transactions, but they can also be targeted by fraudulent activity. This paper proposes a machine-learning-based Fraud Transaction Detection system that classifies transaction records as legitimate or potentially fraudulent using features available in a documented dataset. The proposed workflow includes data validation, preprocessing, handling class imbalance, model training, threshold selection, and evaluation. Candidate algorithms include Logistic Regression, Decision Tree, Random Forest, and Isolation Forest where appropriate to the available labels. Because fraudulent transactions are often rare compared with legitimate transactions, accuracy alone may be misleading; precision, recall, F1-score, the confusion matrix, and precision-recall analysis should be considered. This is a proposed academic system, and no measured accuracy or operational fraud-prevention effectiveness is claimed. A real deployment would require security controls, privacy protection, human review, and monitoring for changing fraud patterns.
Music Recommendation System
Digital music platforms provide millions of tracks, making it difficult for listeners to discover songs that match their interests, current mood, and listening context. A music recommendation system addresses this information-overload problem by ranking tracks likely to be relevant to an individual user. This paper presents a proposed hybrid recommendation system combining content-based filtering and collaborative filtering. Content-based features represent songs using metadata such as genre, artist, language, tempo, and audio descriptors, while collaborative filtering uses user–track interactions such as plays, likes, skips, and playlist additions. A ranking component combines these signals and applies basic constraints to reduce repetition and improve diversity. The proposed architecture includes data collection, preprocessing, feature representation, candidate generation, ranking, feedback capture, and a user interface. The methodology describes baseline models and evaluation using Precision@K, Recall@K, NDCG@K, catalog coverage, and user feedback. No experimental performance results are claimed because the proposed system has not been implemented and tested as part of this paper. Privacy, cold-start handling, popularity bias, and transparency are discussed as key design concerns.
Face Recognition system
Face recognition is a computer vision technique that identifies or verifies a person by analyzing facial characteristics in an image or video frame. It is used in applications such as access management, attendance support, and identity verification, but it also raises important privacy, fairness, and security considerations. This paper presents a conceptual face recognition system that captures an image, detects and aligns a face, extracts a numerical representation, and compares that representation with authorized reference templates. The system can operate in verification mode, where a claimed identity is checked, or identification mode, where a face is compared against a permitted gallery. Performance depends on image quality, lighting, pose, the selected model, and the quality of enrolled reference data. The proposed design includes a controlled matching threshold, secure template handling, and a review path for uncertain matches. This is a student research-paper sample; implementation details, dataset sources, experimental results, and evaluation values should be completed using the author's actual system.
AI-BASED EMAIL SPAM AND SCAM DETECTION
Scholarly manuscript entitled 'AI-BASED EMAIL SPAM AND SCAM DETECTION'.
AI-Based Cyber Threats Detection System
Scholarly manuscript entitled 'AI-Based Cyber Threats Detection System'.
Student Performance Analysis Using Python-Based Data Analytics and Visualization
Scholarly manuscript entitled 'Student Performance Analysis Using Python-Based Data Analytics and Visualization'.
Al- Based Air Quality Monitoring
Scholarly manuscript entitled 'Al- Based Air Quality Monitoring'.
Phishing Email Detection Using NLP
Scholarly manuscript entitled 'Phishing Email Detection Using NLP'.
Ai-based emergency blood doner matching and recommendation sydtem
Scholarly manuscript entitled 'Ai-based emergency blood doner matching and recommendation sydtem' presenting substantial scientific and empirical contributions.
vehicle number plate recognition
Scholarly manuscript entitled 'vehicle number plate recognition'.
AI-BASED FAKE NEWS DETECTION USING MACHINE LEARNING
Scholarly manuscript entitled 'AI-BASED FAKE NEWS DETECTION USING MACHINE LEARNING'.
Auto-Associative Neural Network: Architecture, Learning Process and Applications
Scholarly manuscript entitled 'Auto_Associative_Neural_Network_Base_Paper'.
AI Based questions generation system
Scholarly manuscript entitled 'AI Based questions generation system'.
Intelligent College Chatbot using GPT
Scholarly manuscript entitled 'Intelligent College Chatbot using GPT'.
AI-Based Prediction of Student Placement Readiness
Scholarly manuscript entitled 'AI-Based Prediction of Student Placement Readiness' presenting substantial scientific and empirical contributions.
Healthcare AI
Scholarly manuscript entitled 'Healthcare AI'.
AUTO-ASSOCIATIVE NEURAL NETWORKS
Scholarly manuscript entitled 'AUTO-ASSOCIATIVE NEURAL NETWORKS'.
NLP-Driven Misinformation Identification in Online News Content
The growth of online news platforms and social media has increased the speed at which information reaches the public. Alongside legitimate reporting, misleading, manipulated, and fabricated content can circulate widely before it is verified. Manual verification is valuable but can be time-consuming when large volumes of articles must be reviewed. This paper proposes an NLP-driven approach for identifying news content that resembles examples of misinformation in a labelled dataset. The proposed workflow includes text validation, preprocessing, feature extraction, supervised classification, result presentation, and human review. TF-IDF with Logistic Regression or Multinomial Naive Bayes is considered as a transparent baseline, while contextual embeddings may be explored for capturing semantic relationships. The evaluation plan includes precision, recall, F1-score, accuracy, confusion-matrix analysis, and tests across different sources or time periods. No experimental results are claimed because the proposed system has not been tested in a controlled experiment. The approach treats automated predictions as screening signals rather than definitive proof, since language patterns alone cannot establish factual truth. The design emphasises dataset quality, transparent limitations, and responsible use.
AI Resume Analyzer using Generative AI
Recruitment teams and job seekers often face difficulty matching resumes with job descriptions because resumes use varied formats, terminology, and descriptions of skills. This paper proposes an AI Resume Analyzer using Generative AI to extract resume information, compare candidate profiles with job requirements, identify relevant skills, and generate understandable feedback. The proposed workflow combines document parsing, text preprocessing, skill and experience extraction, semantic similarity, and a generative model that produces a structured analysis. The system is intended to support—not replace—human decision-making by explaining matches, highlighting missing or unclear information, and suggesting resume improvements. A prototype may use PDF/DOCX parsing, embeddings, a vector similarity method, and a controlled language-model prompt. Evaluation should examine extraction quality, relevance of recommendations, consistency, latency, and fairness across varied resume formats. No experimental performance results are claimed because the proposed system has not been benchmarked. The design emphasises consent, privacy, transparency, and avoidance of unsupported or discriminatory hiring conclusions.
Understanding_Cloud_Computing_Vishakha_Shegade_6_Pages.pdf
Scholarly manuscript entitled 'Understanding_Cloud_Computing_Vishakha_Shegade_6_Pages.pdf'.
Cloud Computing and Databases
Cloud computing has changed how organizations provision computing resources and operate data-intensive applications. Databases hosted in cloud environments can scale with demand, support geographically distributed users, and reduce the need to maintain physical infrastructure. At the same time, cloud database design introduces choices around relational and non-relational data models, consistency, availability, latency, cost, security, and regulatory obligations. This review explains core cloud-computing service and deployment models, compares common database approaches, and outlines an end-to-end process for selecting, deploying, and operating cloud databases. It emphasizes that no single architecture is optimal for every workload: effective designs begin with workload requirements and make trade-offs explicit. The paper concludes with practical recommendations for reliability, security, cost control, and future research.
Agentic_AI_Security_Privacy_Challenges.
Scholarly manuscript entitled 'Agentic_AI_Security_Privacy_Challenges.'.
“Predictive Data Analytics for Early Detection of Hidden Business Risks Using Machine Learning.”
Scholarly manuscript entitled '“Predictive Data Analytics for Early Detection of Hidden Business Risks Using Machine Learning.”' presenting substantial scientific and empirical contributions.
AI-BASED COLLEGE INFORMATION ASSISTANT USING RAG
Scholarly manuscript entitled 'AI-BASED COLLEGE INFORMATION ASSISTANT USING RAG'.
Restaurant Recommendation using AI
Scholarly manuscript entitled 'Restaurant Recommendation using AI'.
AI-Based EV Battery Health Prediction with Explainable AI and Database Integration
Scholarly manuscript entitled 'AI-Based EV Battery Health Prediction with Explainable AI and Database Integration'.
AI-BASED SMART IRRIGATION SYSTEM
Scholarly manuscript entitled 'AI-BASED SMART IRRIGATION SYSTEM'.
ONLINE VOTING SYSTEM
The Online Voting System is a Java-based application designed to conduct elections securely and efficiently over the internet. It reduces reliance on traditional paper-based voting by allowing registered users to cast votes online. The system uses Java with MySQL as the back-end database and JDBC for database connectivity. Its core modules include administrator management, voter registration, secure login, voting, and result generation. Administrators can create elections, manage candidates, monitor votes, and view results. Registered voters authenticate before voting, and the system is designed to restrict each voter to one vote per election. The system aims to improve transparency, reduce manual errors, and speed up result processing. It may be useful for colleges, institutions, societies, and organizations conducting internal elections.
SMART WATER QUALITY MONITORING SYSTEM
Water quality monitoring is essential for public health, environmental protection, agriculture, and industrial operations. Conventional monitoring often depends on periodic sample collection and laboratory analysis, which can create delays between a change in water conditions and the response to it. This paper proposes a Smart Water Quality Monitoring System that combines water-quality sensors, a microcontroller, communication connectivity, a data-storage service, and a dashboard to support frequent measurements and timely alerts. The proposed prototype focuses on parameters such as pH, turbidity, and total dissolved solids (TDS), with optional temperature sensing. Readings are validated, time-stamped, compared with configured thresholds, and presented as trends. The system is intended as a screening and decision-support tool; it does not replace certified laboratory testing or establish potability on its own. The paper describes the architecture, workflow, implementation plan, evaluation metrics, potential applications, and limitations. A conceptual trend chart is included to demonstrate how sensor readings may be visualised; it is illustrative and is not presented as collected experimental data.
Smart Parking Availability and Waiting-Time Prediction with Automatic Slot Reallocation
Finding an available parking space is a common problem in crowded areas. Drivers often waste time and fuel while searching for parking spaces. This paper proposes a smart parking approach that combines the Internet of Things (IoT) and Machine Learning (ML) to improve parking management. IoT sensors are used to identify occupied and available parking slots, while a simple Machine Learning model is proposed to predict parking availability and estimate waiting time. The proposed approach also introduces automatic slot reallocation, in which a reserved parking slot can be released and assigned to another waiting user if the original user does not arrive within the specified time. This approach a ims to improve parking space utilization, reduce unnecessary waiting, and provide a better parking experience. The effectiveness of the proposed approach can be evaluated in future work.
Machine Learning-Based Intelligent Water Management System for Agriculture
Scholarly manuscript entitled 'Machine Learning-Based Intelligent Water Management System for Agriculture'.
Agentic AI: Autonomous Agents and Multi-Agent Security
Agentic artificial intelligence (AI) systems combine language models, memory, planning, and tools to pursue goals with limited step -by-step human direction. Multi -agent systems distribute work among specialized agents that cooperate through messages, delegation, and shared context. These capabilities support complex automation but also introduce security risks, including prompt injection, excessive tool permissions, sensitive -data leakage, memory poisoning, insecure inter -agent communication, and cascading failures. This paper reviews agentic AI architecture and proposes a layered security framework based on identity, least privilege, policy -controlled tool use, validated agent communication, protected memory, monitoring, bounded autonomy, and human approval for high -impact actions. It also presents a practical evaluation plan using adversarial test cases and measurable security and utility metrics. This work is a conceptual design and literature -based review; experimental results are not claimed.
Cybersecurity Awareness Among College Students
Scholarly manuscript entitled 'Cybersecurity Awareness Among College Students' presenting substantial scientific and empirical contributions.
Autonomous AI Agents for Software Development:
Scholarly manuscript entitled 'Autonomous AI Agents for Software Development:'.
REAL-TIME SIGN LANGUAGE RECOGNITION SYSTEM
Communication between deaf or hard-of-hearing people and people unfamiliar with sign language can be challenging. This paper presents a proposed Real-Time Sign Language Recognition System that uses a camera to capture hand gestures and computer-vision techniques to classify signs and display corresponding text. The system is organized into video acquisition, hand-region processing, feature extraction, gesture classification, and text-output modules. A practical implementation may use Python, OpenCV, MediaPipe for hand-landmark detection, and a trained machine-learning classifier. The interface is intended to provide immediate visual feedback and can be extended to speech output. Recognition performance depends on the sign vocabulary, lighting, camera quality, signing speed, and diversity of the training data. Evaluation should use a documented dataset and measured test results; no experimental results are claimed in this sample paper.
PostgreSQL Indexing Strategies for Query Optimization
Indexes are data structures that help PostgreSQL locate rows without scanning an entire table for every query. Carefully selected indexes can reduce disk access, accelerate filtering and joins, support ordering, and enforce uniqueness. However, indexes consume storage, add work to inserts and updates, and can become ineffective when their structure does not match the query or data distribution. This paper reviews PostgreSQL's major index types—including B-tree, Hash, GIN, GiST, SP-GiST, and BRIN—along with composite, partial, expression, and covering indexes. It explains how to use EXPLAIN and EXPLAIN ANALYZE, statistics, and workload evidence to evaluate index choices. The central recommendation is to design indexes around real query patterns, measure before and after changes, and balance read performance against write overhead and maintenance cost.
Machine Learning for Software Defect Prediction
Software defects reduce reliability, increase maintenance costs, and can cause service outages or security vulnerabilities. Software defect prediction (SDP) uses historical development data—such as source-code metrics, change history, complexity, and developer activity—to estimate which files, modules, or changes are more likely to contain defects. This paper reviews the typical machine-learning workflow for SDP, compares common algorithms, discusses datasets and evaluation methods, and summarizes deployment risks. Logistic regression, decision trees, random forests, support vector machines, gradient boosting, and neural networks are considered alongside unsupervised and just-in-time approaches. A central finding from the literature is that predictive performance depends strongly on data quality, project context, class imbalance, and evaluation design; results from random cross-validation may overstate performance when future releases or new projects are the real target. The paper recommends time-aware validation, leakage prevention, interpretable risk estimates, and human-in-the-loop prioritization rather than treating model outputs as proof that code is defective.
unused medicines donations and proper handling system.
Unused medicines accumulated in households represent a problem involving medicine wastage, inappropriate storage, and unsafe disposal. Existing studies have examined household accumulation, disposal practices, expiry-related issues, behavioural interventions, and digital donation platforms, but these aspects are generally addressed separately. This paper proposes an integrated mobile application and web portal for responsible management of unused medicines. The proposed system connects individual donors with potential recipients, NGOs, and participating pharmacies through medicine listings and location-based discovery. It records medicine name, quantity, expiry date, condition, image, and location, while providing expiry warnings and handling and disposal guidance. A contribution-point mechanism is also introduced to recognize responsible participation through completed donations and related activities. The paper presents the system architecture, workflow, database design, matching process, safety-oriented controls, and reward mechanism. The current work is primarily a software-system design and implementation proposal; therefore, no clinical or large-scale user-study results are claimed. The expected contribution is a practical digital framework that combines donation coordination, awareness, and responsible participation in a single platform.
Robustness-Aware QAOA Feature Selection for Adversarially Resilient Deepfake Detection
Deepfake detection systems must remain reliable when manipulated media are intentionally modified to evade detection. The base framework considered in this study addresses redundant features, class imbalance, and interpretability through QAOA-based feature selection, quantum-inspired attention, balanced focal loss, and Grad-CAM. However, adversarial robustness remains an open direction. This paper proposes robustness-aware QAOA feature selection, where facial features are selected using both discriminative relevance and stability under controlled perturbations. The selected subset is then processed by the existing quantum-inspired attention mechanism. The work defines the system architecture, research gap, robustness-aware objective, experimental design, evaluation measures, and future validation plan. No experimental improvement is claimed at this stage.
Agentic Retrieval-Augmented Generation (RAG): An Adaptive Framework for Evidence-Grounded Question Answering
Retrieval-Augmented Generation (RAG) connects a large language model (LLM) to external knowledge sources so that responses can be based on evidence beyond the model’s parameters. Conventional RAG pipelines commonly follow a mostly fixed sequence: retrieve a small set of passages, place them in a prompt, and generate an answer. This approach can be inadequate for complex questions that require multiple searches, tool use, source comparison, or correction when the first retrieval is weak. This paper proposes a modular Agentic RAG framework in which specialized agents coordinate query planning, retrieval, tool selection, evidence grading, answer generation, and self-reflection. The system maintains a bounded feedback loop: if retrieved evidence is insufficient, the planner revises the query or selects an additional source; if evidence is adequate, a generator produces a response with citations and a confidence assessment. The paper synthesizes relevant work on RAG, reasoning-and-acting agents, self-reflection, corrective retrieval, and graph-based retrieval. It also presents an evaluation framework covering answer quality, retrieval and grounding quality, efficiency, and ablation tests. The included diagrams illustrate the proposed architecture, the contrast with single-pass RAG, and the evaluation workflow. Because no implementation or benchmark run is reported here, performance charts are explicitly conceptual and the evaluation plan is a proposal rather than a claim of measured improvement. The contribution is a practical, auditable design blueprint for building and evaluating Agentic RAG systems.
Blockchain-Based Secure Academic Credential Verification
Academic credential fraud, manual verification delays, fragmented institutional databases, and limited interoperability motivate trustworthy digital credential systems. This paper proposes a permissioned blockchain architecture for issuing, storing, and verifying academic credentials while minimizing exposure of personally identifiable information. The design combines digital signatures, cryptographic hashes, a permissioned ledger, encrypted off-chain storage, issuer authorization, and an explicit revocation lifecycle. The paper presents the complete issuer–holder–verifier workflow, threat model, privacy considerations, implementation architecture, experimental methodology, evaluation metrics, security analysis, limitations, and future scope. Numerical experimental results are intentionally not fabricated; the evaluation section defines a reproducible procedure for obtaining them.
Privacy_Preserving_ML_Sensitive_Data
Machine learning increasingly relies on sensitive data, creating risks of unauthorized disclosure, inference, memorization, and misuse. This paper proposes a layered privacy-preserving machine-learning framework combining federated learning, differential privacy, secure aggregation, and optional encrypted computation. Raw records remain at participating organizations while protected model updates are aggregated for global learning. The paper presents the threat model, formal privacy concepts, system architecture, training workflow, attack analysis, experimental setup, utility and privacy metrics, comparative analysis, limitations, and future research directions. Numerical experimental results are intentionally not fabricated; the paper defines a reproducible evaluation procedure for generating valid measurements.
Deep Learning-Based IoT Anomaly Detection
The scale and heterogeneity of Internet-of-Things deployments create a large attack surface and generate complex network behavior. This paper proposes a deep-learning anomaly-detection framework that learns representations of IoT traffic and identifies deviations from expected behavior. The design uses preprocessing, temporal representation learning, anomaly scoring, and an alert aggregation layer. Evaluation is defined around precision, recall, F1-score, false-positive rate, and detection latency. Illustrative performance values are synthetic and included only as a formatting example.
An AI-Based Personalized Learning System
Personalized learning requires more than predicting whether a learner will pass or fail. A useful learning system must connect performance evidence to topic-level diagnosis, an appropriate level of challenge, and recommendations that a learner can understand and act upon. This paper proposes an AI-based personalized learning framework that combines student-performance prediction, topic-wise weakness detection, recommendation rules, and an explainability layer. The framework is designed to integrate previous examination marks, quiz scores, assignment performance, attendance, number of attempts, learning-activity completion, and time-on-task when these variables are available. For empirical benchmarking, the verified UCI Student Performance dataset is selected as a reproducible public baseline. The UCI repository reports 649 instances and 33 attributes; the dataset supports classification and regression and contains final grade G3 together with prior-period grades and school, demographic, social, and study-related variables. Because the public dataset does not contain genuine LMS event logs or topic-tagged assessment records, it is used only for the performance-prediction component. Topic-level recommendation is specified as a deployable framework requiring richer course/LMS data. Five supervised models—Logistic Regression, Decision Tree, Random Forest, Gradient Boosting, and Support Vector Machine—are proposed for controlled comparison using stratified cross-validation and Accuracy, Precision, Recall, F1-score, ROC-AUC, and confusion matrices. Recommendation quality is to be evaluated separately with Precision@K, Recall@K, and NDCG@K after implementation. No experimental accuracy or recommendation score is reported in this manuscript because the dataset could not be downloaded into the execution environment and no completed experiment is being represented as performed. The main contribution is therefore a reproducible research design that explicitly links prediction to actionable, explainable learning recommendations while identifying the data requirements and evaluation boundaries of the public benchmark.
AI-Based Hospital Management System: A Modular Framework for Operational Intelligence and Decision Support
Hospitals manage interconnected activities including patient registration, appointment scheduling, clinical documentation, laboratory coordination, pharmacy, bed allocation, inventory, and billing. Fragmented information systems can make these workflows difficult to coordinate and can limit timely operational decisions. This paper proposes a modular AI-based hospital management framework that combines a conventional hospital information system with carefully scoped machine-learning services. The proposed architecture includes a data integration layer, role-based workflow modules, AI services for appointment and resource forecasting, operational anomaly detection, and decision-support interfaces. The design emphasizes that AI outputs should assist—not replace—qualified staff, particularly where patient care is involved. A reproducible evaluation plan is specified using workflow completion time, scheduling utilization, forecast error, alert precision/recall, system latency, and user acceptance measures. Since no prototype deployment or controlled experiment has been performed for this manuscript, all outcome metrics are marked “To be measured”; no performance improvements are claimed. The paper discusses interoperability, data quality, privacy, bias, explainability, cybersecurity, and implementation constraints. The contribution is a system design and evaluation blueprint that can guide subsequent implementation and empirical study.