Artificial Intelligence In Defence Systems: A Review Of Applications, Enabling Infrastructure And Responsible Adoption In The Indian Context
Authors: Sayyad Mohd Zeeshaan Mohd Isaq, Prof. S. R. Borhade, Prof. M. C. Kolse, Prof. S. K. Said
Abstract: Artificial intelligence is reshaping modern defence by strengthening situational awareness, decision-making and operational readiness. This paper develops an integrated review of AI adoption across defence operations, with attention to the Indian context. It examines how machine learning, deep learning, computer vision, natural language processing and big data analytics support intelligence, surveillance and reconnaissance through analysis of satellite, radar and drone imagery; command and control through data fusion and decision support; cyber defence through intrusion detection; logistics and predictive maintenance of equipment; simulation-based training; and autonomous and unmanned platforms. The paper discusses the organizational and infrastructural foundations required for successful implementation, including reliable data, secure interoperable networks, computing facilities, skilled personnel and collaboration among the armed forces, research organizations and industry. It also reviews risks such as adversarial manipulation of AI models, brittleness in unfamiliar conditions, accountability for autonomous functions and international debates on meaningful human control, together with published ethical principles. A five-layer defence AI architecture is proposed with research gaps, objectives, a comparison of applications and challenges. The review concludes that AI should augment, not replace, human judgement in defence.
DOI:
Machines That Predict Their Own Failure: A Review Of Self-Predictive Maintenance Using Sensors, Machine Learning And Digital Twins
Authors: Inamdar Mohammad Hadi Mohammad Shaker, Prof. C. A. Kolse, Prof. M. C. Kolse, Prof. S. K. Said
Abstract: Industrial operations increasingly demand reliable, uninterrupted machines, yet conventional maintenance depends on fixed schedules or on repairing faults after they occur, causing unexpected breakdowns, downtime and high costs. Self-predictive machines monitor their own health and forecast failures early enough to plan maintenance. This paper reviews the technologies that make such machines possible. It covers condition monitoring with vibration, temperature, current and acoustic sensors, signal processing and feature extraction, anomaly detection with autoencoders, fault diagnosis with machine learning and deep networks, and prognostics that estimate remaining useful life with recurrent and convolutional models. Digital twin technology, which maintains a virtual replica of the physical machine updated with live data, is examined as a way to simulate degradation and test maintenance decisions. Benchmark datasets for bearings and turbofan engines are discussed. A five-layer self-predictive machine architecture comprising sensing, edge processing, health analytics, digital twin and maintenance decision layers is proposed, with research gaps, objectives, a comparison of methods and challenges such as scarce failure data, changing operating conditions and explainability. The review concludes that combining sensing, learning and digital twins enables condition-based maintenance that reduces downtime and cost.
AI In Accident Prediction: A Review Of Crash Risk Modelling, Spatio-Temporal Deep Learning And Dashcam Accident Anticipation
Authors: Shingote Atharva Rajendra, Prof. S. R. Borhade, Prof. M. C. Kolse, Prof. S. K. Said
Abstract: Road traffic crashes kill more than a million people each year, and India records one of the highest numbers of road deaths in the world. Predicting where, when and how accidents are likely to occur allows authorities to deploy enforcement and engineering measures, and allows vehicles to warn drivers before a collision. This paper reviews artificial intelligence for accident prediction at two scales. At the network scale, it examines statistical crash-frequency models, machine learning comparisons with traditional regression, and spatio-temporal deep learning that combines historical accidents, traffic, weather, road features and points of interest to estimate accident risk for each region and time. At the vehicle scale, it reviews accident anticipation from dashcam video, in which recurrent and relational models attend to road agents and estimate the probability of an imminent collision several seconds in advance. Datasets, evaluation metrics and explainability methods for identifying risk factors are discussed. A five-layer architecture comprising data sources, data integration, risk modelling, explanation and intervention layers is proposed, together with research gaps, objectives, a comparison of approaches and challenges such as under-reporting, class imbalance, data sharing and false warnings. The review concludes that AI-based accident prediction, linked to concrete interventions, can contribute significantly to road safety.
Cybersecurity For Protecting Your Digital Life: A Review Of Personal Threats, Defensive Technologies And User Awareness
Authors: Gholap Bhagwati Sudhir, Prof. C. A. Kolse, Prof. M. C. Kolse, Prof. S. K. Said
Abstract: Digital communication, online banking, shopping, education and social networking are now part of everyday life, and with this dependence comes exposure to phishing, malware, ransomware, identity theft, password attacks and unauthorized access to personal information. This paper reviews cybersecurity from the perspective of the individual user. It examines the most common threats to personal digital life and the technical and behavioural defences that counter them, including strong password management, multi-factor and passwordless authentication, encryption, firewalls, antivirus and endpoint protection, secure browsing, software updates, backups and phishing detection. Research on how attackers obtain credentials, why users fall for phishing and how experts differ from non-experts in their security habits is surveyed to explain why technology alone is insufficient. A layered personal security architecture is proposed that combines identity protection, device security, network security, data protection and user awareness under continuous monitoring. A research gap analysis, objectives, a comparison of authentication methods and a discussion of challenges are presented, with reference to the NIST Cybersecurity Framework and Indian incident-reporting guidance. The review concludes that a small set of well-chosen habits, supported by usable security tools, can greatly reduce the risk of data breaches, financial fraud and identity theft.
AI-Based Indian Sign Language Recognition Using Computer Vision: A Review Of Hand Tracking, Feature Learning And Gesture Classification
Authors: Andhale Aniket Pandhari, Prof. S. K. Said, Prof. M. C. Kolse
Abstract: Indian Sign Language (ISL) is a primary means of communication for many Deaf and hard-of-hearing people, yet few hearing people understand it, creating everyday barriers in education, healthcare and public services. This paper reviews AI-based sign language recognition with a focus on isolated ISL gesture recognition from ordinary camera input. It traces the field from early wearable and colour-glove systems to modern vision pipelines that use real-time hand and pose landmark extraction, convolutional and recurrent networks, and three-dimensional video models. Key datasets, including INCLUDE for ISL, WLASL for word-level American Sign Language and RWTH-PHOENIX for continuous recognition, are discussed alongside the difference between isolated recognition, continuous recognition and full translation. A five-layer architecture is proposed that captures video, preprocesses frames, extracts MediaPipe hand landmarks, classifies gestures with a lightweight neural network and outputs text and speech. The design is presented as a proposed system; no implementation results or accuracy claims are made. Research gaps, objectives, a comparison of feature approaches and challenges such as signer variation, lighting, two-hand occlusion and limited ISL data are analysed.
Explainable Artificial Intelligence (XAI) In Self-Driving Cars: A Review Of Techniques For Transparent Perception, Prediction And Decision-Making
Authors: Kohakade Vaishnavi Sanjay, Prof. M. C. Kolse, Prof. S. K. Said
Abstract: Self-driving cars use cameras, radar, LiDAR and deep learning to detect objects, recognize traffic signs, identify pedestrians and make driving decisions without direct human control. Most of these models behave as black boxes, making it difficult to understand why a vehicle braked, changed lanes or failed to detect an obstacle. This lack of transparency creates problems for safety assurance, debugging, user trust, accountability and regulation. Explainable Artificial Intelligence (XAI) addresses these problems by providing understandable reasons for AI decisions. This paper reviews XAI for autonomous driving across the perception, prediction and planning stages. It examines post-hoc methods such as LIME, SHAP and gradient-based saliency maps, visual attention and VisualBackProp for end-to-end steering models, textual explanations that describe driving actions in natural language, object-induced action explanations, and inherently interpretable designs. Stakeholder needs of developers, passengers, safety assessors and regulators are analysed. A layered XAI-enabled autonomous driving architecture is proposed, and research gaps, objectives and a comparison of explanation techniques are presented. The review concludes that integrating explainability throughout the driving stack, and evaluating explanations for faithfulness and usefulness, is essential for safe, reliable and trustworthy autonomous transportation.
Artificial Intelligence In The Food Industry: A Review Of Quality Inspection, Spoilage Prediction, Demand Forecasting And Traceability
Authors: Galande Atharva Satish, Prof. N. G. Rokade, Prof. M. C. Kolse, Prof. S. K. Said
Abstract: The food industry must produce safe, fresh and affordable food for a growing population while reducing waste and cost, yet roughly one-third of food produced is lost or wasted. Many operations such as sorting, inspection, storage monitoring and demand planning are still performed manually, which makes them slow, inconsistent and dependent on human experience. This paper reviews how Artificial Intelligence supports food producers, processors, retailers and restaurants. It examines computer vision and convolutional neural networks for quality grading and defect inspection, electronic noses and IoT sensors combined with machine learning for spoilage detection and shelf-life prediction, time-series forecasting for demand planning, and digital traceability, including blockchain pilots, for food safety. Classical neural-network applications in food science and recent deep learning reviews are surveyed. A five-layer AI food management architecture with image capture, sensor monitoring, analytics modules, alerts and a traceability dashboard is proposed, with research gaps, objectives, a comparison of techniques and a discussion of challenges including variable natural products, data scarcity, hygiene constraints and cost for small enterprises. The review concludes that AI can improve consistency, reduce waste and strengthen food safety across the supply chain.
Emotion Recognition Using Artificial Intelligence: A Review Of Facial, Speech And Multimodal Affective Computing
Authors: Kashid Aditi Eknath, Prof. M. C. Kolse, Prof. S. K. Said
Abstract: Emotion recognition enables computers to infer human emotional states such as happiness, sadness, anger, fear, surprise and neutrality from facial expressions, voice and other behavioural signals. It underpins affective computing applications in healthcare, education, customer service, driver monitoring and human–computer interaction. This paper reviews AI-based emotion recognition. It introduces categorical and dimensional models of emotion, then surveys facial expression recognition from hand-crafted features to deep convolutional networks trained on laboratory and in-the-wild datasets, speech emotion recognition from acoustic features and deep models, and multimodal fusion that combines face, voice and text. Benchmark datasets including CK+, FER-2013, AffectNet and RAVDESS are discussed. Scientific and ethical debates are examined, particularly evidence that facial movements do not map reliably onto inner emotions across people and contexts, along with privacy and consent concerns. A five-layer architecture comprising input capture, preprocessing, feature learning, fusion and classification, and application with feedback is proposed, together with research gaps, objectives and a comparison of modalities. The review concludes that emotion recognition can make interaction more natural when used transparently, with consent and with awareness of its limits.
Smart Street Light Fault Identification For Multiple Smart Streets: A Review Of IoT Monitoring, Anomaly Detection And Maintenance Automation
Authors: Kale Kunal Sunil, Prof. M. C. Kolse, Prof. S. K. Said
Abstract: Street lighting is essential for urban safety and accounts for a significant part of municipal electricity use. Traditional maintenance depends on manual night patrols or public complaints, so faulty lamps may stay dark for days, wasting energy and creating safety risks. This paper reviews technologies for a smart street light fault identification system that covers many streets. It examines IoT nodes that measure current, voltage, power and light intensity at each pole, wireless networks including ZigBee mesh and LoRaWAN for city-wide coverage, and control platforms that switch and dim lamps. Fault detection methods are reviewed, from threshold rules on electrical parameters to unsupervised anomaly detection that identifies abnormal behaviour such as flickering, partial failure, cable faults and power theft. Smart lighting deployments, India's national LED street-lighting programme and the cybersecurity risks of networked lighting are discussed. A five-layer architecture comprising pole-level sensing, communication, a central monitoring platform with fault analytics, a maintenance workflow and a city dashboard is proposed, with research gaps, objectives, a comparison of communication technologies and challenges such as cost, interoperability and security. The review concludes that automatic fault identification reduces downtime, maintenance cost and energy use while improving public safety. Index Terms—Smart street lighting, IoT, fault detection, anomaly detection, ZigBee, LoRaWAN, smart city, predictive maintenance
Deepfake Detection Technology: A Review Of Lightweight Transfer-Learning Detectors, Artefact Cues And Generalization To Unseen Manipulations
Authors: Thorat Sanika Santosh, Prof. M. C. Kolse, Prof. S. S. Bhor, Prof. S. K. Said
Abstract: Deepfake technology uses deep learning to create or manipulate images and videos so that a person appears to say or do something that never happened. Although it has legitimate uses, it can spread misinformation, enable fraud and damage reputations. This paper reviews deepfake detection technology with a focus on lightweight, deployable detectors and the generalization problem. It examines how manipulated faces are produced, the benchmark datasets used for training and testing, and detectors based on transfer learning, in which convolutional networks such as MobileNetV2 and EfficientNet pretrained on ImageNet are fine-tuned to classify face images or video frames as real or fake. It reviews artefact cues that detectors exploit, including blending boundaries, frequency-domain traces of up-sampling and fine-grained texture differences captured by attention, and the finding that detectors trained on one generator can transfer to others when trained with suitable augmentation. Detection of images from diffusion models is also discussed. A five-stage architecture comprising media input, face extraction and preprocessing, transfer-learning feature extraction, classification and a user interface built with a lightweight web framework is proposed, together with research gaps, objectives, a comparison of detector backbones and challenges such as compression, dataset bias and adversarial evasion. The review concludes that predictions should be treated as indications rather than proof and that robust generalization remains the main open problem.
AI-Based Smart Assistants: A Review Of Natural Language Understanding, Task Automation And Multimodal Interaction
Authors: Dhokare Apeksha Chimaji, Prof. N. G. Rokade, Prof. M. C. Kolse, Prof. S. K. Sai
Abstract: Smart assistants help users complete everyday tasks such as answering questions, searching information, setting reminders, checking the weather, opening applications and controlling devices through natural voice or text commands. This paper reviews the technologies that make such assistants possible and the way they have evolved from rule-based chatbots to systems built on large language models that can call external tools. The review covers input processing through speech recognition and text normalization, natural language understanding with intent classification and slot filling, dialogue management, task execution through application programming interfaces and operating-system commands, and response generation through text and speech synthesis. Retrieval-augmented generation and tool-use methods that ground assistant answers in current information are examined, along with personalization and memory. A modular architecture suitable for implementation in Python is proposed, covering the interaction, understanding, reasoning, skills and integration layers. A research gap analysis, objectives, a comparison of assistant design paradigms and a discussion of challenges in accuracy, connectivity, privacy and multilingual support are presented. The review concludes that combining a reliable task-oriented core with a language-model reasoning layer offers an efficient, user-friendly and extensible assistant for education, healthcare, business and home automation.
Artificial Intelligence Of Things (AIoT): A Review Of Architectures, Learning Techniques And Applications Of IoT With AI
Authors: Buchake Rameshwari Hanumant, Prof. M. A. Auti, Prof. M. C. Kolse, Prof. S. K. Said
Abstract: The Internet of Things (IoT) connects sensors, cameras, smartphones and machines to the internet so that they can collect and exchange data, while Artificial Intelligence (AI) analyses that data to recognize patterns and make decisions. Their combination, often called the Artificial Intelligence of Things (AIoT), enables systems that monitor, control and manage activities with minimal human involvement. This paper reviews AIoT from the perspectives of architecture, intelligence and application. It explains layered IoT architectures, communication protocols and context-aware computing, then surveys machine learning and deep learning for IoT data analytics, including streaming analytics, anomaly detection and edge-based inference, as well as federated learning for privacy. Applications in smart homes, smart cities, healthcare, agriculture and industrial IoT are reviewed. Security threats such as large IoT botnets are discussed as a key concern. A five-layer AIoT architecture comprising perception, connectivity, edge intelligence, cloud analytics and application layers is proposed, along with research gaps, objectives, a comparison of learning approaches and an analysis of challenges in interoperability, energy, privacy and security. The review concludes that AIoT can greatly improve automation, accuracy and productivity when built with security and privacy in mind.
Predicting Natural Disasters Using Artificial Intelligence And Machine Learning: A Review Of Methods, Data Sources And Early Warning Architectures
Authors: Bhalerao Shruti Santosh, Prof. M. A. Auti, Prof. M. C. Kolse, Prof. S. K. Said
Abstract: Floods, earthquakes, cyclones, landslides and droughts cause heavy loss of life and property, and their impact depends strongly on how early communities are warned. Conventional physics-based forecasting remains essential, but it struggles to exploit the volume and variety of data now produced by satellites, ground sensors and historical archives. This paper reviews how Artificial Intelligence (AI) and Machine Learning (ML) are being used to estimate disaster risk and support early warning. It surveys data sources, including rainfall, river levels, seismic waveforms, reanalysis fields and satellite imagery, and examines models ranging from random forests and support vector machines for susceptibility mapping to long short-term memory networks for flood forecasting, convolutional and transformer networks for earthquake detection, and graph-based models for medium-range weather prediction. A five-layer architecture for an AI-assisted early warning system is proposed, linking data ingestion, preprocessing, hazard-specific models, risk fusion and alert dissemination. The paper identifies research gaps such as data scarcity in ungauged regions, limited uncertainty reporting and the weak connection between forecasts and community action. It concludes that AI predictions should be treated as decision support that complements, rather than replaces, physical models and expert judgement.
Graph Neural Networks For Recommendation Systems: A Review Of Architectures, Training Objectives And Open Challenges
Authors: Alisha Raju Shaikh, Prof. S. K. Said, Prof. M. C. Kolse
Abstract: Recommendation systems help users discover products, films and articles on digital platforms. Classical collaborative and content-based filtering struggle to capture non-linear relationships and high-order connectivity between users and items, especially when interaction data are sparse. Graph Neural Networks (GNNs) address this by modelling interactions as a graph in which users and items are nodes and interactions or attributes are edges, and by learning embeddings through message passing over neighbourhoods. This paper reviews GNN-based recommendation. It introduces graph convolution, graph attention and inductive sampling, then surveys landmark recommenders including graph convolutional matrix completion, PinSage, Neural Graph Collaborative Filtering, LightGCN, knowledge-graph attention networks and session-based GNNs. Training objectives such as Bayesian personalized ranking and evaluation with recall and normalized discounted cumulative gain are explained. A layered architecture covering graph construction, embedding propagation, prediction and serving is proposed, together with research gaps, objectives and a comparison of models. Challenges of over-smoothing, scalability, cold start, fairness and explainability are discussed. The review concludes that GNNs provide a principled way to exploit relational structure in recommendation, with simpler propagation often outperforming heavier designs.
AI-Powered Voice Assistants: A Review Of Speech Recognition, Language Understanding, Speech Synthesis And Security
Authors: Khebade Anushka Santosh, Prof. M. C. Kolse, Prof. S. K. Said
Abstract: An AI-powered voice assistant lets people interact with computers and devices through natural speech rather than keyboard and mouse. It recognizes spoken commands, understands the user's intent, performs an action and replies with synthesized speech. This paper reviews the end-to-end speech pipeline that makes voice assistants possible. It covers wake-word detection with small-footprint neural networks, automatic speech recognition from early deep acoustic models to recurrent, transformer-based and self-supervised systems trained on large and diverse audio, spoken language understanding through intent classification and slot filling, dialogue and action execution, and neural text-to-speech that produces natural voices. Evaluation measures such as word error rate and intent accuracy are explained, and the security of voice interfaces against hidden and inaudible commands is examined. A five-layer architecture comprising audio capture, speech recognition, language understanding, action execution and speech output is proposed, along with a research gap analysis, objectives and a comparison of speech recognition approaches. Accessibility benefits for students, professionals and people with disabilities are discussed. The review concludes that modern speech and language models enable simple, hands-free and efficient human–computer interaction, provided that accuracy for diverse accents, privacy and voice security are addressed.
Wireless Network Security: A Review Of Threats, Security Protocols From WEP To WPA3, And AI-Assisted Intrusion Detection
Authors: Lokhande Sanghmitra Vinayak, Prof. R. V. Varhade, Prof. M. C. Kolse, Prof. S. K. Said
Abstract: Wireless technologies such as Wi-Fi, Bluetooth and mobile networks provide convenient connectivity without cables, but because signals travel through the air they are exposed to eavesdropping, unauthorized access, spoofing, denial-of-service and man-in-the-middle attacks. This paper reviews wireless network security. It analyses the main threats and traces the evolution of Wi-Fi security protocols: the cryptographic failures of Wired Equivalent Privacy, the improvements of Wi-Fi Protected Access and WPA2 with IEEE 802.11i, attacks such as key reinstallation that exposed flaws in WPA2's handshake, and the stronger but still scrutinized WPA3 with simultaneous authentication of equals. Weaknesses in Bluetooth key negotiation and in LTE link layers are also examined. Defensive measures, including strong passphrases, enterprise authentication with 802.1X, network segmentation, firewalls, firmware updates and wireless intrusion detection using machine learning on public datasets, are reviewed. A five-layer wireless security architecture comprising device hardening, secure association, encrypted transport, monitoring and incident response is proposed, along with research gaps, objectives, a comparison of protocols and challenges such as legacy devices, IoT constraints and user behaviour. The review concludes that layered defence with modern protocols and continuous monitoring is essential.
Big Data Analytics Using Artificial Intelligence: A Review Of Platforms, Learning Techniques And Applications
Authors: Kharade Shruti Sharad, Prof. M. C. Kolse, Prof. S. K. Said
Abstract: Organizations in healthcare, banking, education, retail and government generate data in volumes, varieties and speeds that traditional database tools cannot process effectively. Big Data Analytics using Artificial Intelligence (AI) combines scalable data platforms with machine learning and deep learning to identify patterns, predict future trends and generate actionable insights from this data. This paper reviews the field from both the systems and the learning perspective. It explains the characteristics of big data, the evolution of processing frameworks from MapReduce and the Hadoop distributed file system to in-memory engines such as Apache Spark, stream processors such as Apache Kafka and Flink, and the lakehouse model. It then examines how machine learning and deep learning are scaled through distributed libraries and frameworks, and how analytics serves applications such as disease prediction, fraud detection, recommendation systems and learning analytics. A five-layer architecture comprising data sources, ingestion, storage, processing and AI, and visualization is proposed, followed by a research gap analysis, objectives, a comparison of frameworks and a discussion of challenges including data quality, privacy, cost and the hidden maintenance burden of machine-learning systems. The review concludes that AI-driven big data analytics converts raw data into timely, meaningful information for better decision-making.
AI-Based Fake News Detection Using Machine Learning: A Review Of Multilingual Detection, Automated Fact-Checking And Messaging-Platform Misinformation
Authors: Pawar Tanushri Sarjeet, Prof. M. R. Kumbhoje, Prof. M. C. Kolse, Prof. S. K. Said
Abstract: Misinformation spreads not only through websites and social networks but also through messaging platforms and in many languages, which is especially relevant in India. This paper reviews AI-based fake news detection using machine learning with emphasis on three aspects that are less covered by English-only text classifiers: multilingual detection, automated fact-checking and misinformation on messaging platforms. It discusses the social science of fake news and the characteristic features of false articles, such as sensational titles and simpler language, and reviews machine learning classifiers trained on labelled datasets. Automated fact-checking pipelines that identify check-worthy claims, retrieve evidence and verify claims against it are examined, together with cross-lingual transformer models and resources for Indian languages. Studies of misinformation on messaging platforms and pandemic-related fake news datasets are discussed. A five-layer architecture comprising multilingual ingestion, preprocessing, classification, evidence-based verification and user-facing explanation is proposed, along with research gaps, objectives, a comparison of approaches and challenges such as low-resource languages, encrypted messaging, evolving narratives and freedom-of-expression concerns. The review concludes that combining machine learning classification with evidence-based fact-checking provides more trustworthy detection.
Diffusion Models For Image Generation: A Review Of Denoising Probabilistic Models, Latent Diffusion, Guidance And Controllable Synthesis
Authors: Sayyed Rida Fatima Sharif, Prof. S. R. Borhade, Prof. M. C. Kolse, Prof. S. K. Said
Abstract: Diffusion models are generative models that create high-quality, realistic images by learning to reverse a gradual noising process. This paper reviews the working principles, architectures and applications of diffusion models for image generation. It explains the forward process, which adds Gaussian noise to training images over many steps, and the reverse process, in which a neural network, usually a U-Net, learns to predict and remove noise. The review traces the field from its thermodynamic origins to denoising diffusion probabilistic models (DDPM), the score-based stochastic differential equation view, and faster samplers such as DDIM. It examines classifier and classifier-free guidance, latent diffusion models that operate in a compressed latent space for efficiency, text-to-image systems conditioned on language embeddings, and spatial control methods that add edges, poses or depth as conditions. Applications in digital art, image editing, restoration, medical imaging and animation are discussed. A five-stage architecture is presented, along with evaluation metrics, research gaps, objectives, a comparison with GANs and VAEs and challenges such as slow sampling, computational cost, bias, copyright and misuse for deepfakes. The review concludes that diffusion models offer an excellent balance of quality and diversity, with efficiency and responsible use as the main open problems.
File Integrity Verification Using Cryptographic Hashing: A Review Of Hash Functions, Keyed Hashes, Merkle Trees And Integrity Monitoring Systems
Authors: Patel Faheem Faruk, Prof. J. M. Patade, Prof. M. C. Kolse, Prof. S. K. Said
Abstract: Files on computers, servers and cloud storage can be altered by malware, intruders, faulty hardware or insiders, often without visible signs. File integrity verification detects such changes by computing a cryptographic fingerprint of each file and comparing it with a trusted reference. This paper reviews the cryptographic foundations and system designs for file integrity verification. It explains the properties of cryptographic hash functions, namely preimage, second-preimage and collision resistance, and examines the families in use, including the deprecated MD5 and SHA-1, which suffered practical collisions, the SHA-2 family and faster modern designs such as BLAKE2. Keyed-hash message authentication codes and digital signatures are reviewed as means of protecting stored hashes from attackers who could otherwise recompute them. Merkle hash trees for efficient verification of large file collections, file integrity monitoring tools, kernel-level integrity measurement and transparency logs are discussed. A five-layer architecture comprising file discovery, hashing, protected baseline storage, scheduled and event-driven verification, and alerting with response is proposed, together with research gaps, objectives, a comparison of hash algorithms and challenges such as legitimate updates, performance on large files and baseline tampering. The review concludes that strong hashes combined with protected baselines provide reliable tamper detection.
