Daily Archives: November 14, 2026

Authors: Gaikwad Harsh Ajay, Prof. N. G. Rokade, Prof. M. C. Kolse, Prof. S. K. Said

Abstract: Autonomous vehicles use artificial intelligence to understand their surroundings, decide how to move and control steering, acceleration and braking with minimal or no human intervention. This paper reviews the AI technologies across the full autonomous driving stack. It traces the field from the DARPA challenge vehicles to modern systems and explains the levels of driving automation. Perception with cameras, radar and LiDAR, including deep object detection and point-cloud networks, is examined, followed by localization and mapping through LiDAR odometry and simultaneous localization and mapping, motion planning and control, and learning-based approaches including direct perception, conditional imitation learning, deep reinforcement learning and end-to-end planning-oriented networks. Large-scale datasets and simulation for training and validation are discussed. A five-layer modular architecture for an autonomous vehicle is proposed, together with research gaps, objectives, a comparison of modular and end-to-end approaches and an analysis of challenges including safety validation, rare scenarios, adverse weather, mixed traffic, cybersecurity, ethics and regulation. The review concludes that AI has made autonomous driving technically feasible in defined conditions, but that broad deployment depends on demonstrable safety and robust handling of the unexpected.

DOI: https://doi.org/10.5281/zenodo.24220422

Published by: vikaspatanker

Authors: Durafe Sejal Dattatray, Prof. N. G. Rokade, Prof. M. C. Kolse, Prof. S. K. Said

Abstract: Fake news and misinformation on digital platforms threaten public trust and information integrity, and manual verification cannot keep pace with the volume of content created daily. This paper reviews AI-based fake news detection beyond simple text classification, focusing on deep learning and social-context methods. It examines early credibility features on social media, recurrent networks that model the time series of user responses to rumours, hybrid models that combine article text with user behaviour, explainable models that use user comments to justify decisions, and graph-based geometric deep learning that analyses how news spreads. Machine-generated fake news and multimodal datasets combining text and images are also discussed, together with support vector machines and random forests as baselines. A five-layer real-time detection architecture is proposed that fuses content, user and propagation signals, followed by research gaps, objectives, a comparison of approaches and a discussion of challenges including early detection before content spreads, cross-platform generalization, adversarial manipulation and the need for explainability. The review concludes that combining what is said, who says it and how it spreads produces more accurate and trustworthy fake news detection than content analysis alone.

DOI: https://doi.org/10.5281/zenodo.24220335

Published by: vikaspatanker

Authors: Doke Shubham Somnath, Prof. N. G. Rokade, Prof. M. C. Kolse, Prof. S. K. Said

Abstract: Healthcare systems face a shortage of trained doctors and specialists, heavy hospital workloads, rapidly growing volumes of medical data, diagnostic delays and unequal access to quality care, particularly in rural areas. Artificial Intelligence (AI), including machine learning, deep learning, computer vision and natural language processing, is increasingly used to learn patterns from medical data and support clinicians in diagnosis, treatment planning and patient monitoring. This paper reviews AI in healthcare through a three-stage collect–predict–care pipeline. In the collection stage, medical images such as X-ray, CT, MRI, retinal and skin photographs, laboratory results, electronic health records and wearable readings are gathered and prepared. In the prediction stage, convolutional neural networks analyse images, machine learning models estimate patient risk from records and language models interpret clinical notes. In the care stage, results are presented to doctors as decision support, and the doctor makes the final diagnosis and treatment decision. The review surveys landmark studies for each stage, proposes a layered clinical decision support architecture, and discusses challenges of privacy, bias, explainability, infrastructure cost and regulation. It concludes that AI can make healthcare faster, more accurate and more accessible when it is validated in real hospitals and always operates under clinical supervision.

DOI: https://doi.org/10.5281/zenodo.24220253

Published by: vikaspatanker

Authors: Dhole Siddharth Dnyaneshwar, Prof. N. G. Rokade, Prof. M. C. Kolse, Prof. S. K. Said

Abstract: Plant diseases, unpredictable weather and soil problems cause major yield losses every year, and farmers in many regions still depend on visual inspection or delayed expert visits to identify diseases. Late detection often leads to excessive use of chemical sprays, which raises costs and harms the environment. This paper reviews AI-powered systems that detect crop diseases early from leaf images and provide smart recommendations for treatment and crop management. The review examines image datasets such as PlantVillage and PlantDoc, convolutional neural networks and transfer learning for disease classification, object detection models for locating lesions in field images, and lightweight networks for offline smartphone use. It also covers how detection results can be combined with weather, soil and crop-stage information to recommend targeted, dose-appropriate interventions and preventive practices. A five-layer architecture comprising data acquisition, preprocessing, a detection engine, a recommendation engine and a farmer-facing application is proposed. The paper identifies research gaps such as the drop in accuracy from laboratory to field images, limited regional crop coverage and the lack of explainable advice, and it concludes that a multilingual, mobile-first assistant combining disease detection with context-aware recommendations can reduce losses and chemical overuse.

DOI: https://doi.org/10.5281/zenodo.24220185

Published by: vikaspatanker

Authors: Bhor Anwit Rajendra, Prof. M. A. Auti, Prof. M. C. Kolse, Prof. S. K. Said

Abstract: Sudden rain and intense sunlight affect daily commuting, yet people frequently leave home without an umbrella or forget it at offices, buses and shops. Conventional umbrellas offer protection but no intelligence. This paper reviews the design of a smart umbrella reminder system that combines Internet of Things (IoT) sensing, wireless proximity tracking and machine-learning-based weather prediction to remind users at the right moment. The review covers environmental sensing of rainfall, temperature, humidity and light; Bluetooth Low Energy (BLE) links between the umbrella and a smartphone for left-behind alerts based on received signal strength; consumption of weather forecasts and short-term precipitation nowcasts; and context-aware logic that considers the user's schedule and location before issuing a reminder. Low-power hardware design, including duty cycling and energy harvesting, is examined because the device must operate for months on a small battery. A layered architecture covering the umbrella node, the smartphone gateway, a cloud service and the user interface is proposed, followed by a research gap analysis, objectives, comparison of connectivity and prediction options, and a discussion of challenges. The review concludes that a low-cost, context-aware smart umbrella is a practical example of how everyday objects can become helpful IoT companions.

DOI: https://doi.org/10.5281/zenodo.24220115

Published by: vikaspatanker

Authors: Bangar Piyush Rajesh, Prof. S. K. Said, Prof. M. C. Kolse

Abstract: Social media, websites and online portals have become major sources of news, but they also allow false and misleading information to spread rapidly, creating confusion and social harm. Fake news detection aims to identify such content automatically so that users and platforms can respond quickly. This paper reviews machine learning and natural language processing approaches to detecting fake news from the text of articles and headlines. It covers how fake news is defined and why it spreads, benchmark datasets such as LIAR, ISOT and FakeNewsNet, preprocessing steps including cleaning, tokenization and stop-word removal, feature extraction with bag-of-words, n-grams and TF-IDF, and classification with Logistic Regression, Naive Bayes and Support Vector Machines, as well as deep models based on BERT. A practical five-layer pipeline that can be implemented with Python libraries such as Pandas, NumPy, Scikit-learn and NLTK is proposed, together with research gaps, objectives and a comparison of algorithms. Limitations of text-only detection, dataset bias and the need for explainable, regularly updated models are discussed. The review concludes that a well-built text-classification pipeline is an effective first line of defence and a strong foundation for richer detection systems.

DOI: https://doi.org/10.5281/zenodo.24220040

Published by: vikaspatanker

Authors: Chapekar Ved Sudhir, Prof. M. A. Auti, Prof. M. C. Kolse, Prof. S. K. Said

Abstract: Fraud in banking, digital payments, online shopping and insurance causes large financial losses and erodes customer trust. Traditional fraud detection relies on predefined rules and manual verification, which struggle with complex and constantly changing fraud patterns. Artificial Intelligence (AI) enables systems that analyse large volumes of transaction data in real time, recognize unusual behaviour and generate alerts. This paper reviews AI-based fraud detection. It explains the characteristics of fraud data, including extreme class imbalance, concept drift and delayed labels, and surveys statistical approaches, supervised classifiers such as logistic regression, random forests and gradient boosting, unsupervised anomaly detection, sequence models that learn from customers' transaction histories, and graph neural networks that expose fraud rings. Hybrid systems that combine supervised and unsupervised learning, benchmark datasets and evaluation metrics suited to imbalanced problems are discussed. A five-layer real-time fraud detection architecture is proposed, along with research gaps, objectives, a comparison of methods and challenges including false positives, explainability and adversarial adaptation by fraudsters. The review concludes that layered AI systems with human investigators in the loop provide the most effective defence.

DOI: https://doi.org/10.5281/zenodo.24219829

Published by: vikaspatanker

Authors: Chakave Sakshi Revannath, Prof. M. A. Auti, Prof. M. C. Kolse, Prof. S. K. Said

Abstract: The number and sophistication of cyberattacks, including malware, phishing, ransomware and denial-of-service campaigns, continue to grow, and signature-based security tools struggle to recognize new or modified threats. Machine Learning (ML) offers a complementary approach that learns the characteristics of normal and malicious behaviour from data and flags suspicious activity in real time. This paper reviews ML-based threat detection with emphasis on network intrusion detection. It traces the evolution from the classical intrusion-detection model to rule-based systems, and examines supervised classifiers such as Decision Trees, Random Forests and Support Vector Machines, unsupervised anomaly detectors and deep learning models. Public benchmark datasets, feature engineering from network flows, evaluation metrics and the practical difficulties of class imbalance, concept drift and false alarms are discussed. A layered architecture for a real-time detection system is proposed, comprising traffic capture, feature extraction, a hybrid detection engine, alert management and response integration. A research gap analysis, objectives and a comparative analysis of algorithms are presented. The review concludes that ensemble ML models combined with explainable alerts and continuous retraining can significantly improve detection accuracy and reduce false alarms in operational security environments.

DOI: https://doi.org/10.5281/zenodo.24219742

Published by: vikaspatanker

Authors: Bacche Sanket Kailas, Prof. S. K. Said, Prof. M. C. Kolse

Abstract: Managing personal and organizational expenses with paper receipts and spreadsheets is slow, error-prone and offers little insight into spending behaviour. Cloud-based expense trackers address these problems by centralizing financial records, synchronizing them across devices and applying analytics to reveal patterns. This paper reviews the technologies behind such systems. It examines cloud service and deployment models, intelligent capture of receipts through Optical Character Recognition (OCR) and document-understanding models, automatic categorization of transactions, budget forecasting and anomaly detection, and integration with banking data through consented interfaces such as India's Account Aggregator framework. Security is treated as a core requirement: the review discusses encryption in transit and at rest, multi-factor authentication, role based access control for multi-user organizations and common web-application vulnerabilities. A five-layer reference architecture is proposed that separates client applications, an API gateway, microservices, data storage and an analytics engine. A research gap analysis, objectives, comparison of storage and capture options and a discussion of challenges are presented. The review concludes that combining automated data capture with explainable analytics and strong security controls can make expense tracking both effortless and trustworthy for individuals and enterprises.

DOI: https://doi.org/10.5281/zenodo.24219665

Published by: vikaspatanker

Authors: Dhole Om Santosh, Prof. N. G. Rokade, Prof. M. C. Kolse, Prof. S. K. Said

Abstract: Artificial Intelligence is moving from single models that answer one question at a time toward agentic systems that plan, use tools, coordinate subtasks and execute multi-step workflows. A Multi-Agent AI System (MAAS) contains two or more autonomous or semi-autonomous agents that interact to achieve a shared or related goal, where each agent may have a specialized role, memory, tools and decision process. This paper reviews the foundations and recent advances of such systems. It traces the classical theory of multi-agent systems, including agent communication languages and the contract-net protocol, and examines how large language models have revived the field through frameworks for role-playing, conversation-driven collaboration, standardized operating procedures and debate. Coordination topologies such as orchestrator–worker, hierarchical, peer-to-peer and blackboard designs are compared, along with mechanisms for memory, tool access and verification. A layered reference architecture is proposed that separates user interface, orchestration, specialized agents, shared memory and tools, and governance. The paper also analyses failure modes, including error propagation, coordination overhead, prompt injection and cost, and identifies research gaps in evaluation and reliability. It concludes that multi-agent designs are most valuable when tasks decompose naturally and when intermediate outputs are verified.

DOI: https://doi.org/10.5281/zenodo.24219567

Published by: vikaspatanker