Cloud-Based Intelligent Replenishment System Using Machine Learning and Business Analytics
Authors: Dr. Pankaj Malik, Bhumi Wadhwani, Pushkar Rathore, Miraj Chourasiya, Stuti Bhati, Pushkar Bhati
Abstract: Efficient inventory replenishment is essential for minimizing stock-outs, excess inventory, and operational costs in modern retail and supply-chain environments. Traditional replenishment approaches generally rely on fixed reorder points and historical averages, which may not adequately capture nonlinear demand patterns, seasonal variations, promotional effects, and changing lead times. This paper proposes a Cloud-Based Intelligent Replenishment System (CBIRS) that integrates machine learning, cloud computing, and business analytics to enable dynamic and data-driven inventory replenishment. The proposed framework utilizes cloud-based data ingestion and storage to integrate historical sales, product, pricing, promotion, inventory, seasonal, and supplier lead-time information. Multiple machine learning models, including Random Forest, XGBoost, LightGBM, CatBoost, and Long Short-Term Memory (LSTM), are evaluated for demand forecasting, followed by a hybrid forecasting model for intelligent replenishment decisions. The predicted demand is integrated with safety-stock estimation, reorder-point calculation, lead-time analysis, and dynamic order-quantity optimization. Experimental evaluation demonstrates that the proposed hybrid model achieves an MAE of 9.72, RMSE of 15.63, MAPE of 7.86%, and R² of 0.94, outperforming the individual baseline models. At the inventory-management level, the proposed CBIRS reduces the stock-out rate to 4.2% and overstock rate to 8.3%, while achieving an estimated 21.1% reduction in total inventory cost compared with the static replenishment approach and a 95.8% service level. The results indicate that integrating machine learning forecasting with cloud-based analytics and dynamic replenishment can significantly improve inventory availability, cost efficiency, and supply-chain decision-making.
CAMSD-Net For SIDD-Small Real-World Image Denoising Using Lightweight Multi-Scale Deep Learning Framework
Authors: Arpit Urmaliya, Jitendra Kumar Maitra
Abstract: Effective image denoising is a challenging task for real-world smartphone images due to the complex noise resulting from the limited response of the sensors and the varying illumination conditions. The paper proposes and tests a light weight deep learning framework called Convolu-tional Adaptive Multi-Scale Denoising Network CAMSD-Net to restore high quality image from noisy smartphone photos in the meanwhile retaining the structure and visual quality. The pro-posed methodology employs the SIDD-Small sRGB dataset with 160 pairs of noisy and clean images which are systematically validated, with 70% used for training, 15% for validation, and 15% for testing, and with random 256×256 image patches extracted from each image, leading to a fixed 384 paired image patches used for testing. The network is implemented based on multi-scale convolutional network, local spatial learning, dilated contextual feature extraction, feature fusion and skip connections, and is trained for 30 epochs with the Adam optimizer and learning rate of 0.001. Experimental results show the superior performance of CAMSD-Net V2 over classical denoising schemes and the first version of the network. The proposed model achieves the highest Peak Signal-to-Noise ratio (PSNR) of 36.5564 dB, and the Structural Similarity Index (SSIM) of 0.925337, and minimizes the Mean Squared Error (MSE) of 0.00032221, Root Mean Squared Error (RMSE) of 0.016248, and Mean Absolute Error (MAE) of 0.012848. In addition, the proposed framework is effective in real world smartphone image restoration as evidenced by its performance on 356/384 test patches (92.71%) with respect to the improvement of PSNR.
DOI: http://doi.org/10.5281/zenodo.22829914
Self-Learning Artificial Intelligence for Adaptive Data Stream Analytics: Comparative Evaluation of Robustness, Scalability, and Predictive Reliability
Authors: Deepshikha Rajpoot, Sonam Chamber, Poonam Bhartiya
Abstract: With the evolution of digital transformation landscape, today data stream analytics has become an essential paradigm involving real-time decision-making in various domains such as finance, healthcare and Internet of Things (IoT). Nonetheless, the fact that data streams are non-stationary–concept drift, noise or change pattern are some of the difficulties for traditional static ma-chine learning models. In this research, we provided a comprehensive meta-analysis of the existing self-learning AI architec-tures to analyze adaptive data streams. This study compares the performance of these models using findings from recent litera-ture across three key criteria: robustness to environmental change, scalability in high-throughput scenarios and prediction accu-racy over longer time horizons. This meta-analysis combines information from 30+ landmark studies by evaluating Adaptive Random Forests, Evolving Fuzzy Systems, and Self-Adjusting Memory (SAM) models. The experiments show that deep learning-based adaptive systems achieve better predictive reliability, while we found that some instance-based learning models (i.e., SAM-kNN) perform well on various heterogeneous drift types without retraining effort and hyperparameter tuning. In addition, we discover a general trade-off between computational scalability and adaptivity granularity. The approach used is that of a systematic review, with quantifiable measures to evaluate architectural performance. In this paper, we further discuss the impact of delayed labeling and noise interference on the stability of adaptive learners. We conclude this paper with the implica-tions of these findings in building explainable and energy-efficient AI systems, along with a few future research directions merging federated stream learning with generative AI. This review provides a practical starting point for researchers and practi-tioners that aim to deploy resilient, autonomous and scalable AI systems in dynamic and high-throughput data environments by bridging the gap between theoretical frameworks and deployment challenges.
Artificial Intelligence-Driven Recommendations And Their Impact On Consumer Decision-Making In E-Commerce
Authors: Dr. V. Vetrivel, Dr. K. Vinayagam, Dr. A. Gokulakrishnan, Dr. P. Sasikumar
Abstract: The rapid growth of e-commerce has changed how consumers search for products, compare alternatives and make purchasing decisions. Artificial Intelligence (AI)-driven recommendation systems have become an important component of digital commerce because they analyse consum-er preferences and behavioural patterns to present relevant products, services and content. This study examines the impact of AI-driven recommendations on consumer decision-making in e-commerce, with particular focus on recommendation relevance, perceived usefulness, conven-ience and purchase intention. This study adopts a descriptive, quantitative research design. Fol-lowing the structure of the supplied model paper, the study uses an illustrative sample of 200 online shoppers. Data will be collected through a structured questionnaire using a five-point Likert scale. Descriptive statistics, Chi-square and one-way ANOVA are employed to examine relationships and differences. The illustrative results show a significant association between fre-quency of exposure to AI recommendations and purchase intention. The results also indicate significant differences in consumer decision-making across different levels of perceived recom-mendation usefulness. The study suggests that e-commerce businesses should improve recom-mendation accuracy, provide transparent AI-based suggestions, protect consumer data, reduce irrelevant recommendations and maintain effective customer support. The study concludes that AI-driven recommendations can positively influence online consumer decision-making when they are relevant, useful, convenient and trustworthy.
DOI: https://zenodo.org/records/23008829
Role of Entrepreneurship Development Clubs in Developing Entrepreneurial Mindset Among Higher Secondary Students in Kerala
Authors: Sivadasan K, Dr Bakkialakshmi S
Abstract: Purpose: Entrepreneurship Development Clubs (EDCs), promoted by the Department of Indus-tries and Commerce, Government of Kerala, function in a large number of schools across the State to inculcate an entrepreneurial culture among the youth. This study examines the role of EDCs in developing the entrepreneurial mindset namely risk-taking ability, innovative and crea-tive thinking, self-confidence, decision-making skill and problem-solving orientation — among higher secondary students in Kerala. Design/Methodology/Approach: This descriptive study uses primary data collected from 240 higher secondary students who are active members of EDCs functioning in four selected schools of Kerala, through a structured questionnaire. Percentage analysis, descriptive statistics and Analysis of Variance (ANOVA) have been used to analyse the data. Findings: The study found that the level of entrepreneurial mindset among student members increases significantly with the duration of EDC membership, while the stream of study makes little difference except for the decision-making dimension.
DOI: https://zenodo.org/records/23009802
Experimental Study On Standard And High-Strength Concrete Incorporating Waste Tyre Rubber And Silica Fume
Authors: Umesh Seetha, Dr. Ala Satyanarayana
Abstract: This study investigates M40 standard-grade and M70 high-strength concrete incorporating waste tyre rubber as partial replacement of natural fine and coarse aggregates together with silica fume. The experimental programme evaluates workability, compressive strength, split tensile strength, flexural strength, stress-strain behaviour, Young's modulus, rebound number, ultrasonic pulse velocity, sorptivity and resistance to 5% MgSO4, HCl and H2SO4 solutions. The mixes include control concrete, concrete with 4% silica fume, concrete with 4% coarse waste tyre rubber and silica fume, and concrete containing both fine and coarse waste tyre rubber. The study reports 264 specimens, with 132 specimens for each concrete grade. Results show that increasing tyre-rubber content reduces workability and mechanical strength because of weak bonding between rubber particles and cement paste. The target mean strength was obtained for rubberized concrete con-taining 4% coarse rubber and 4% silica fume, while the required target mean strength was not achieved for M70 rubberized concrete. Non-destructive, sorptivity and chemical-exposure results provide additional assessment of quality and durability.
AI – Driven Predictive Maintenance For Industrial Rotating Machinery Using Vibration Analysis
Authors: Dr. Biswajit Datta, Rahila Begum Gadi
Abstract: Industrial rotary machines experiencing unplanned outages owing to bearing and gear defects cost economies and safety greatly. In this paper, we introduce a hybrid deep learning algorithm for predictive maintenance through the combination of 1-Dimensional Convolutional Neural Net-works (1D-CNN) and Bi-directional Long Short-Term Memory (BiLSTM) networks in order to perform diagnosis as well as Remaining Useful Life (RUL) estimation using vibration data. The introduced algorithm uses multi-scale feature learning and temporal dependency modeling in order to detect degradation trends without the need for much feature engineering. Evaluation experiments conducted on the CWRU bearing dataset illustrate high classification accuracy even when the vibration data contains varying amounts of noise and low RUL prediction error. Com-parative experiments show the effectiveness of the introduced hybrid model when compared to other models such as SVM, Random Forest, CNN, and LSTM.
DOI: http://doi.org/10.5281/zenodo.23156475
Digital Financial Inclusion and GST Formalization among Street Vendors in India: The Role of UPI and Artificial Intelligence
Authors: Assistant Professor Dr. Prathima V, Professor Dr. N Nagaraja
Abstract: The rapid pace of digitalization, particularly since the implementation of the Goods and Services Tax (GST) and the widespread adoption of Unified Payments Interface (UPI), has brought about a major shift in the nature of the informal business in India. Street vendors are increasingly getting integrated into the formal economy via smartphone payments, QR code payments, mo-bile banking and digital applications, which are largely cash-based and unofficial transactions. This change has given rise to smart vendors; those who actively involve themselves in the digital commerce world and financial systems. The study aims to explore the potential of UPI and artificial intelligence (AI) in promoting digital financial inclusion and GST integration for street vendors and small-scale business owners in India. The study is conceptual and analytical and is based on the latest research on GST, AI, digital payments and financial inclusion, Government publications and literature. It discusses the ease of GST com-pliance with AI tools such as automated bookkeeping, AI chatbots, fraud detection, predictive tax analytics, and smart account-ing apps, as well as the push towards formal taxation by digital payment platforms. The study revealed that numerous obstacles exist, such as lack of knowledge regarding the implementation of GST, lack of digital skills, cyber security concerns and limited access to technology. The findings indicate that both UPI and AI technologies have driven a major rise in financial accessibility and digital transaction engagement, and can potentially make compliance easier, help boost transparency, and minimize filing mistakes.
An Advanced Hybrid Deep Learning-Based Early Rumor Detection Model For Social Media Users
Authors: Reeta Bourasi, Dr. Sakshi Rai
Abstract: The importance of social media posts has been increased in our daily life activities. The wide distribution of different rumors and misinformation has affected many business organizations. The wide spread of rumors on social networking sites has brought many social, economic, personal, and political problems throughout the world. False information distorts public perceptions, undermines corporate reputations, disrupts communities, diminishes governmental and business integrity, and poses a significant threat to societal stability. False information affects public perception, harms corporate reputations, and threatens societal stability. Previous machine learning methods like Random-Forest and Gradient-Boosting have been used to detect rumors. However, the proposed hybrid model employs a Convolutional Neural Network combined with Bidirectional Long Short-Term Memory and Gated Recurrent Unit (CNN-BiLSTM and CNN-BiGRU) for classifying social media posts. This paper proposes an advanced hybrid deep learning-based approach that targets early rumor detection on Twitter and Facebook, achieving an accuracy of 98% – 99%, which surpasses baseline machine learning models.
Test Paper Title By Saquib
Authors: Mohd saquib siddiqui, Ashar Ahmed
Abstract: Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat. Duis aute irure dolor in reprehenderit in voluptate velit esse cillum dolore eu fugiat nulla pariatur. Excepteur sint occaecat cupidatat non proident, sunt in culpa qui officia deserunt mollit anim id est laborum.
DOI: http://doi.org/
Generative AI-Based Framework for Automated Knowledge Discovery from Unstructured Data
Authors: Assistant Professor Ashish Shukla, Assistant Professor Pooja Dwivedi
Abstract: The rapid rise of unstructured text data in various scientific, technical and corporate contexts is driving the necessity for auto-mated tools to distill usable knowledge from this raw text without extensive curatorial effort. In this paper we introduce a framework which uses generative AI techniques to convert unstructured documents into structured knowledge bases that can be queried using a pipeline of large language models combined with ontology-based extraction, entity canonicalization and retriev-al-augmented generation. Our architecture solves three recurring problems of knowledge discovery: extraction hallucinations, entity fragmentation and lack of alignment between queries and evidence in extracted knowledge. Our framework uses a two-step extraction approach that distinguishes between candidate generation and structural validation, with an identity resolution component that unifies various forms of surface-level entities into canonical entities. Experimentally, on a wide range of hetero-geneous corpora, we show that our framework greatly mitigates noise introduced by node generation from unstructured docu-ments, improves consistency of entity resolution and enables multi-hop querying over document-driven knowledge structures.
Generative AI -Based Employee Skill Gap Analysis and Personalized Upskilling Recommendations
Authors: Assistant Professor Dr.M.Sangeetha, Assistant Professor Dr.Banoth Rajendranath
Abstract: Rapid integration of generative artificial intelligence (AI) into enterprise processes highlights critical gaps in workforce readi-ness in terms of skills required for successful implementation of new technologies. This paper introduces a framework which will enable use of generative AI for conducting dynamic skill gap analysis and providing personalized recommendations for upskilling. The framework includes mapping of competencies according to shifting role demands, individual performance as-sessment through AI and adaptation of learning path for each individual based on results. Using the information obtained from surveys about AI adoption in enterprises and competency maps developed in recent research, this paper will show how genera-tive AI can be used for making static learning programs adaptive and relevant to the current needs of learners. The problem discussed in the paper is related to not only technical AI literacy, but also to human skills, such as critical thinking, ethical deci-sion-making and change management. Comparative analysis of methods shows that personalized training offered by generative AI is more engaging and relevant than generic models of training programs.
The Efficacy and User Perception of Digital Health Interventions for Tinnitus: A Mixed-Methods Evaluation
Authors: Abubakar Mairiga, Naveed Anwar, Biddy Casselden
Abstract: Background: Tinnitus is a chronic auditory disorder affecting an estimated 10–15% of the United Kingdom adult population and approximately 50 million Americans, characterised by the perception of phantom sounds in the absence of external stimuli. Despite the absence of a definitive cure, mobile health (mHealth) applications have emerged as accessible self-management tools. However, critical gaps remain in the evaluation of these applications across dimensions of accessibility, evidence-based content, functionality, and personalisation. Objective: This study aimed to conduct a comprehensive, mixed-methods evaluation of mobile applications for tinnitus management, focusing on their efficacy, user satisfaction, and the tension between high over-all positive sentiment and persistent limitations in evidence-based content, personalisation, and user accessibility. Methods: A three-strand mixed-methods design was employed: (1) sentiment analysis of 13,096 user reviews from 42 tinnitus management applications across Google Play and Apple App Store (July–December 2022), collected using Appbot; (2) a cross-sectional online survey of 171 participants with self-reported tinnitus, conducted between March and June 2024; and (3) in-depth semi-structured interviews with 19 purposively selected participants. Quantitative data were analysed using descriptive statistics, chi-square tests of independence, and Spearman rank correlation. Qualitative data were thematically analysed using NVivo. Re-sults: Sentiment analysis revealed that 72.38% of 13,096 reviews were positive, and the overall average app rating was 4.29 out of 5 stars. Google Play exhibited higher positive sentiment (79.49%) compared to iOS (65.42%). ReSound Tinnitus Relief was the top-rated application by volume (n=3,452 reviews; mean rating=4.61). Survey findings revealed that 91.4% of 171 tinnitus sufferers had never used a tinnitus-specific mobile application. Among the 14 active app users, ambient sounds (71.43% useful/very useful) and information about tinnitus (64.29%) were the most valued features. Spearman rank correlation confirmed significant positive associations between app feature information and health content quality (ρ=0.73, P<.001), app use and user experience (ρ=0.68, P<.001), and app use and well-being indicators (ρ=0.71, P<.001). Interview data revealed four core themes: apps as lifesavers for sleep and emotional stress management, value and connectivity through app use, rele-vance of educational content, and effects on tinnitus management and well-being. Conclusions: Tinnitus management mobile applications demonstrate significant promise as tools for sleep improvement, stress reduction, and patient education, with strong positive user sentiment. However, a paradoxical finding, high satisfaction among the 8.6% of users who adopt apps versus 91.4% non-adoption, reveals major systemic barriers including low awareness, inadequate personalisation, and insufficient evidence-based content. Collaborative development among app developers, clinicians, and researchers is essential to close this efficacy–accessibility gap.
Statistical Learning Framework For Predictive Analysis Of Multivariate Time- Series Data
Authors: Dr.R.Pushpalatha, G.RANI PADMINI, V.GOMATHI
Abstract: Multivariate time-series forecasting is subject to certain difficulties, including intricate inter-variable dependency, non-stationary behavior, and temporal dynamics. This paper introduces a new statistical learning model that incorporates decomposed-based pre-processing and adaptive learning of graphs for predictive analytics. The framework utilizes a multi-layered approach that includes three main stages: decomposition of the data into the components such as trend, seasonality, and residual parts; learning the graphs for each of the components based on inter-variable dependency; and, finally, a combination of statistical and neural forecasting methods. Results obtained from benchmark experiments show significant improvement of performance of the proposed framework in comparison with state-of-the-art baseline models. In particular, component-based modeling with learned dependency graphs outperforms other methods consistently for all forecasting horizons in the presence of correlations between channels and seasonality effects.
DOI: https://doi.org/10.5281/zenodo.23531170
Blockchain – Enabled Framework For Enhancing Transparency In Modern E- Commerce Transactions
Authors: S.Priyadarshini, Rashmi D
Abstract: Today's e-commerce systems are affected by the problems associated with information asymmetry, non-transparency of the supply chain, and control of transactions by intermediary parties, which impede trust in the process. In this paper, a blockchain-based framework for increasing the level of transparency in e-commerce transactions is suggested. It combines distributed ledger technologies with smart contract automation and IoT technologies for recording every transaction in an immutable way. The suggested methodology uses a Proof of Authority consensus algorithm tailored for e-commerce needs and a multi-level architecture consisting of product registration, transaction processing, and dispute solving stages. To test the performance of the suggested solution, simulation experiments focused on assessing the latency of the transaction process, dispute resolution, and counterfeiting prevention capabilities have been conducted. Comparison of the results with traditional e-commerce systems shows that using the suggested framework leads to reducing dispute rates and faster transaction verification, preserving the system's scalability at the same time
DOI: https://doi.org/10.5281/zenodo.23531360
Predictive Analytics for Consumer Spending Patterns in Digital Payment Ecosystems
Authors: M.Gokul, K.Jeevalakshmi
Abstract: – Predictive modeling of the parameters affecting water quality is crucial for public health, agricultural sustainability, and proper environmental action in advance. Existing lab-based monitoring techniques, being reliable and accurate, are not only expensive but also limited in terms of spatio-temporal coverage, which reduces their practical applicability. This research work aims to create an effective machine learning model for WQI and WQC prediction using the set of standard physicochemical water parameters. In particular, the methodology incorporates pre-processing using mean imputation and min-max normalization, feature importance analysis, and ensembles of several ML models, including Random Forest, XGBoost, Gradient Boosting, and Multi-layer Perceptron. As demonstrated by experiments, ensembles of tree-based algorithms are generally more accurate than individual baseline approaches in terms of prediction reliability for various types of water bodies. The paper also explores feature dominance and shows that pH, dissolved oxygen, and turbidity have the strongest effect on water quality state.
DOI:
Data-Driven Prediction Of Water Quality Parameters Using Machine Learning Techniques
Authors: M.Gokul, K.Jeevalakshmi
Abstract: Predictive modeling of the parameters affecting water quality is crucial for public health, agricultural sustainability, and proper environmental action in advance. Existing lab-based monitoring techniques, being reliable and accurate, are not only expensive but also limited in terms of spatio-temporal coverage, which reduces their practical applicability. This research work aims to create an effective machine learning model for WQI and WQC prediction using the set of standard physicochemical water parameters. In particular, the methodology incorporates pre-processing using mean imputation and min-max normalization, feature importance analysis, and ensembles of several ML models, including Random Forest, XGBoost, Gradient Boosting, and Multi-layer Perceptron. As demonstrated by experiments, ensembles of tree-based algorithms are generally more accurate than individual baseline approaches in terms of prediction reliability for various types of water bodies. The paper also explores feature dominance and shows that pH, dissolved oxygen, and turbidity have the strongest effect on water quality state.
DOI: https://doi.org/10.5281/zenodo.23557991
Sustainable Synthetic Routes for the Construction of Functionalized Heterocyclic Scaffolds
Authors: Professor Dr Venkateswara Rao Arikatla
Abstract: The preparation of functionalized heterocyclic structures using sustainable approaches has gained great importance within con-temporary synthetic organic chemistry. This review focuses on recent progress in sustainable synthetic approaches to obtain various heterocycles, paying special attention to microwave-assisted chemistry, ultrasound, electroactivation, photoredox cataly-sis, mechanochemistry, and biocatalysis. It is evident from the discussed literature that non-classical activation approaches invariably exceed traditional thermal activation in terms of reaction times, energy and waste production. Quantitative green parameters such as atom economy, E-factors, process mass intensity, and reaction mass efficiency prove that solvent-free or aqueous medium-based multicomponent reactions provide the most sustainable synthetic approach. The employment of hetero-geneous catalysts, including magnetic nanoparticles and ionic liquids, contributes to greater recyclability and lower impact on the environment. Comparing microwave-assisted procedures with other approaches, it becomes clear that these processes sig-nificantly decrease reaction times from hours to minutes while maintaining high yields. Electrochemical and photochemical approaches, on the contrary, provide unique selectivity that cannot be achieved using thermal activation. Sustainable synthetic approaches should further develop in terms of process intensification, flow chemistry, and predictive green metrics.
Investigation of Nanomaterials for High-Efficiency Energy Storage Devices
Authors: Research Scholar Md Javed, Assistant Professor Mr Gorakh Waluba Gangurde
Abstract: The efforts aimed at maximizing the efficiency of energy storage have led to detailed studies of nanoarchitectures of electrodes that address the kinetic and structural limitations of bulk materials. In this regard, the current research systematically explores the preparation, structural features, and electrochemical behavior of nanocomposites of graphene, MXenes, and transition metal oxides for supercapacitor and lithium-ion battery technologies. It was proposed to use the combination of hydrothermal and ex situ oxidative polymerization methods to synthesize Ti₃C₂Tₓ MXenes/polyaniline and reduced graphene oxide/transition metal oxide nanocomposites. Cyclic voltammetry, galvanostatic charge-discharge, and impedance spectroscopy results have shown that MXene/PANI composites provide specific capacitance of more than 430 F g⁻¹ with good cycling stability up to 10,000 cycles, while CoNiP/rGO nanocomposite electrodes demonstrate capacities close to 477 mAh g⁻¹ with retention. Comparison of the studied systems with the reference carbon and oxide ones shows that nanoarchitectures where conductive scaffolds pro-vide resistance to restacking of the structure, while redox-active components provide redox-active sites can be considered a solution.
Emergency Route Optimization During Disasters Using AI And Iot: A Comprehensive Review of Sensing Approaches, Routing Algorithms, Report-ed Evidence, and Deployment Challenges
Authors: Arshad Salim Shaikh, Professor A.P. Banger, Professor S. B. Bhosale, Professor Karad V. A.
Abstract: Disasters such as floods, earthquakes and cyclones damage roads and delay ambulances, fire brigades and rescue teams, while conventional navigation relies on static maps that ignore flooding, blockages and congestion. This paper presents a focused narrative review that synthesises five studies on AI-based emergency and evacuation routing: a decentralized reinforcement-learning framework that couples Dijkstra-style routing with traffic-signal control; a machine-learning recommendation system that predicts congestion for ambulances; a flood-sensor system that converts water-level readings into dynamic edge weights for evacuation routing; a graph-based method that reconfigures road directions in real time; and a hybrid quantum machine-learning approach to escape routing. The review examines three connected themes: sensing approaches and routing models, the strength and comparability of the reported evidence, and the conditions that matter when routing is extended into a full IoT-to-dashboard pipeline. All five studies were assessed from their published abstracts and available excerpts, not read in full; the one headline figure found is a 74% improvement in evacuation efficiency, and it cannot be ranked against the other results, which are mostly qualitative. Only one study links physical sensor readings to routing. The review contributes a comparative table, a conceptual architecture for the proposed system, which is not experimentally validated, and a prioritised agenda of research gaps.
AI Factory Safety Monitor: A Real-Time Computer-Vision System for PPE Compliance, Danger-Zone Intrusion and Machine-Hazard Alerting on Industrial Shop Floors
Authors: Bhor Pranali Dinesh, Professor Bangar A. P., Professor Bhosale S.B, Professor Karad V. A
Abstract: Factory accidents continue to arise from lapses that are simple to describe but hard to police continuously: a worker without a helmet, a hand inside a press envelope, a person in the path of a moving vehicle. Manual supervision and periodic audits cannot watch every station at every moment. This paper presents the AI Factory Safety Monitor (AFSM), an edge-deployed computer-vision system that analyses ordinary industrial camera streams in real time to detect missing personal protective equipment (PPE), intrusions into configurable danger zones, and unsafe proximity to operating machinery. AFSM combines a single-stage object detector, a multi-object tracker, a pose-based posture check and a rule engine that converts raw detections into graded alerts using temporal filtering to suppress false alarms. Alerts are delivered locally (beacon, siren) and remotely (MQTT, SMS, supervisor dashboard), and a compliance-analytics layer records events for audit under occupational health and safety manage-ment practice. We describe the architecture, algorithms, deployment considerations and a proposed evaluation protocol, and discuss privacy, reliability and the limits of vision-based safety as a supplement to, not a replacement for, certified safety-rated controls.
AI-Based Smart Weapon Detection and Threat Assessment in CCTV Surveillance: A Review
Authors: Disha Narayan Kale, Professor Bangar A.P, Professor Bhosale S.B, Dr. A. A. Khatri
Abstract: Closed-circuit television (CCTV) networks now produce far more footage than human operators can watch, and delayed recog-nition of an armed person can cost lives. This paper reviews AI-based approaches to weapon detection and threat assessment in CCTV surveillance. Four recent studies are examined in detail: an edge-deployable weapon and threat detector built on a light-weight CNN and Mask R-CNN, an edge–cloud collaborative framework for real-time object detection, a hierarchical context-tree approach to anomaly detection, and a review of deep-learning benchmarks for video surveillance anomaly detection. The studies are compared on architecture, datasets, hardware, evaluation metrics and reported results. The review shows that object detectors reach high accuracy on curated weapon classes, but small and imbalanced datasets, occlusion, false alarms, limited edge compute and the lack of a context-aware risk-scoring layer still prevent dependable deployment. A conceptual layered pipeline that joins edge detection, location-aware anomaly scoring and cloud-assisted model updates is outlined as a direction for future work.
Smart Farm Fence Monitoring and Intrusion Detection System with AI-Based Location Track-ing: A Review and Conceptual Architecture
Authors: Ghule Shubham Santosh, Professor Bangar Abhishek Popat, Professor S. B. Bhosale, Professor Karad V. A.
Abstract: Crop damage and theft caused by wild animals, stray animals and human intruders remain a persistent problem for farmers, and a passive fence neither detects nor reports a breach. This paper presents a focused narrative review of verified studies on intru-sion detection, edge artificial intelligence (AI) and location tracking for farm protection, giving priority to IEEE journal publica-tions and supplementing them with IEEE conference and open-access sources. Four themes are examined: camera-based AI detection and deterrence; low-cost sensor-triggered and LoRa-based fog or edge systems; GNSS and LoRa-based position tracking of animals; and validation practice. The reviewed work shows that lightweight deep-learning detectors can run on low-power edge devices, that a PIR-and-camera strategy has been reported to save energy, and that signal-strength information can reduce the energy spent on continuous GPS. However, the accessible evidence is dominated by prototype-scale studies, loca-tion tracking is rarely integrated with intrusion detection, and false-alarm rate, detection latency and independent validation are seldom reported. The paper contributes a comparative table of representative studies, a proposed seven-layer conceptual archi-tecture for a smart farm fence that combines IR/PIR sensing, camera-based AI, GPS-collar tracking and geofencing, a leakage-resistant evaluation protocol, and a prioritised list of research gaps. The architecture is conceptual and has not been experimen-tally validated.
Smart Classroom and Timetable Scheduler: A Review of Constraint-Based, Optimization, and Hybrid Scheduling Approaches
Authors: Gunjal Pritam Sandip, Professor Abhishek Popat Bangar, Professor S. B. Bhosale, Professor Karad V. A.
Abstract: Building a college timetable takes far more than filling empty slots. Classes, rooms, teachers and students all have to fit together under firm rules and local preferences, while shared space keeps shrinking and teaching moves between in-person and online modes. This paper looks at five research papers on timetabling: a systematic review of practice in higher education, a mixed-integer programming (MIP) method, a hybrid-teaching model, a comparison of solvers from the International Timetabling Competition 2019 (ITC 2019), and a web-based Smart Classroom and Timetable Scheduler. Between them they draw on con-straint validation, round-robin allocation, integer programming, matheuristics, metaheuristics and explicit modelling of online and in-person delivery. One study describes a two-stage MIP matheuristic that finished second at ITC 2019 and solved 29 of the 30 instances; it did well on small and medium cases but scaled poorly to very large ones. The ITC 2019 overview reports gaps of up to 94.46% against the available lower bounds, with five instances now proven optimal. The practical scheduler reports conflict-free timetables but gives no numerical benchmarks. Overall, the papers suggest that no single method suits every institution. A practical smart scheduler should pair checked input data and strict hard-constraint enforcement with a feasi-bility-first, improvement-second solving strategy, optional hybrid-teaching support, open evaluation and human review. Key open problems are scalability, benchmark coverage, fair evaluation, real-time rescheduling, and the risk of accepting a generated timetable without checking it.
Artificial Intelligence and Internet of Things-Based Food Authenticity and Adulteration Detection in Spices: A Review Paper
Authors: Kurhade Aryan Ram, Professor Bangar Abhishek P, Professor Bhosale Sachin B, Dr. Khatri Anand A.
Abstract: Culinary spices represent one of the most commonly adulterated agricultural commodities worldwide due to the complexity of their supply chain and the high economic value of the product. Traditional laboratory analyses (e.g., High-performance liquid chromatography, Mass spectrometry, and Wet chemical tests) are destructive, time-consuming, and insufficient for the high-throughput screening of spice powders or botanicals. This paper conducts a narrative literature review of 22 selected sources published between 2016 and 2026 that fall under seven themes: spice supply-chain risks and adulteration profiles, computer vision and deep learning approaches, portable Vis–NIR spectroscopy, electronic nose and gas sensing, multi-sensor fusion and chemometric analysis, machine learning methods for quantifying adulteration concentrations, and Edge-IoT hardware solutions. The reviewed articles demonstrate that Convolutional Neural Networks and transfer learning can reach an excellent level of precision in distinguishing between whole spices and botanical intruders, while compact Vis-NIR spectral sensors can detect food dyes and starch fillers in ground powders. SVM/Random Forest classifier-based MOS gas-sensor arrays can detect spoiled spices and vision-spectral fusion can have more accurate detection than a single modality. Nevertheless, most of the data presented in the literature is based on experiments performed on an oversimplified scenario of adulteration, and there are not enough publicly available multi-modal sensor datasets. This research presents a thematic literature review, a comparison of features in fourteen selected papers, a proposed conceptual framework, and recommendations.
Cutoff Trend Prediction for Engineering Admis-sions Using Machine Learning
Authors: Mule Darshan Dipak, Bhosale Sachin Bajirao, Bangar Abhishek Popat
Abstract: Engineering admission in Maharashtra is carried out through merit-based Centralised Admission Process (CAP) rounds, in which every college, branch and reservation category closes at a different cutoff percentile each year. Students usually rely on the previous year's cutoff list to decide their preference order, but cutoffs move from year to year because of changes in seat matrices, examination difficulty, branch demand and category-wise competition. Existing college-predictor tools mostly com-pare a candidate's score with last year's closing val-ues and rarely model how the cutoff itself is changing. This paper reviews five selected studies on admission prediction, cutoff-based college prediction and time-series forecasting, and iden-tifies gaps in temporal modelling, validation and uncertainty reporting. Based on these findings, a Machine Learning Framework for Cutoff Trend Prediction is proposed. The framework con-verts official CAP cutoff lists into a structured dataset of about 239,840 records, derives lag-based and trend features for every college-branch-category-round combination, and trains re-gression models such as Ridge regression, Random Forest and gradient boosting, with LSTM and ARIMA considered as optional time-series baselines. The forecast cutoff for the coming year is compared with a student's percentile to group colleges as Safe, Tar-get or Dream. A pro-totype web application with search, prediction and history modules has been developed to de-liver the results. The framework aims to improve the accuracy, transparency and practical use-fulness of cutoff information for engineer-ing aspirants. Quantitative evaluation of the proposed models is identified as future work.
AI-Based EV Charging Station Load Balancing: A Comprehensive Review of Data Sensing, Machine Learning Methods, and IoT Integration
Authors: Nighot Onkar Kailas, Professor Bangar Abhishek Popat, Professor S. B. Bhosale, Dr. A.A.Khatri
Abstract: The rapid growth of electric vehicles (EVs) concentrates charging demand in time and space, which can overload individual charging stations and local grid assets while neighbouring stations remain underused. Fixed-rule or first-come-first-served control cannot adapt well to uncertain arrivals, prices and grid limits. Artificial intelligence (AI), combined with IoT metering and communication, is increasingly proposed for forecasting charging demand and for balancing load across charging stations. This paper presents a focused narrative review of studies that use machine learning (ML), deep learning (DL) and reinforce-ment learning (RL) for EV charging load forecasting, scheduling and load balancing. It examines three connected themes: data sensing, learning methods, and IoT architectures. The reviewed literature reports LSTM-based short-term load forecasting on Adaptive Charging Network data with a mean absolute error of 4 kW, a graph-based RL approach that balances load across a station network through dynamic pricing in simulation, batch RL that coordinates several stations using real-world charging data, and surveys that classify RL methods into centralized and decentralized categories. However, most studies are simulation-based, single-dataset or single-campus, and few report field validation, communication failure handling, or closed-loop opera-tion on physical chargers. The paper contributes a comparative table of representative studies, a conceptual data-to-decision architecture (not experimentally validated), and a prioritised list of research gaps.
IoT-Based Smart Rock Fall Detection and Early Warning System for Mining and Hilly Regions: A Review
Authors: Nitin Patil, Professor Abhishek Popat Bangar, Professor S. B. Bhosale
Abstract: Rockfalls are a recurring hazard in mining areas, road corridors, and hilly regions. They are difficult to predict because they usually show very little pre-failure deformation and collapse within a short time. Internet of Things (IoT) technology, which combines low-cost field sensors, wireless communication, cloud processing, and automated alerting, offers a practical way to monitor such slopes continuously. This review synthesizes five selected scholarly works that cover vibration-based stability assessment of falling-type rock masses, 5G-IoT monitoring with a tangent-angle warning criterion, micro-seismic rockfall detection with failure forecasting, a multi-sensor rockfall early warning system in the Himalaya, and an IoT-based landslide early warning framework deployed in India. The literature suggests that displacement alone is a weak indicator for brittle rock failure, and that dynamic indicators such as dominant frequency and acceleration amplitude ratio, or event-based indicators such as accumulated seismic energy, can show instability earlier. Field deployments also show that communication reliability, power supply, seasonal environmental effects, and the absence of validated alert thresholds are major practical constraints. The paper compares the studies’ methods and contributions, identifies cross-cutting challenges, and proposes directions for future investi-gation. Because the review is based on five selected papers rather than an exhaustive database search, it is presented as a fo-cused narrative review and not as a systematic review.
AI-Driven Intelligent Judicial Case Management System: A Review
Authors: Parth Vishwas Kale, Professor Bangar A.P, Professor Bhosale S.B., Dr. A. A. Khatri
Abstract: Courts in many jurisdictions, and in India in particular, face a persistent backlog that manual scheduling and file handling can-not clear. This paper reviews five recent studies that together cover the main building blocks of an AI-driven intelligent judicial case management system: rule-based case prioritisation, machine-learning prediction of case delay and duration, large-language-model (LLM) based judgment prediction, AI-assisted trial workflows for criminal cases, and public perception of automation at different stages of adjudication. The studies are compared on objective, data, method, reported results and stated limitations. On this basis a six-layer conceptual architecture is outlined, with explainability and human oversight as core layers. The review suggests that (i) prioritisation and delay prediction offer the most immediate and least contentious gains, (ii) outcome-prediction accuracies are promising but are not directly comparable across studies and are not sufficient for deployment, and (iii) public acceptance depends on keeping the judge in control of the later stages of decision-making. Open challenges include data quality, bias, explainability, legal regulation and evaluation in real courts.
AI-Based Smart Blood Management System: A Comprehensive Review of Data Sources, Machine Learning Methods, and IoT Integration
Authors: Pradyumna Vyanktesh Kulkarni, Professor Bangar A.P, Professor Bhosale S.B, Dr. A. A. Khatri
Abstract: Blood cannot be manufactured, has a short shelf life and must be available at the right place and time, yet many blood banks still rely on manual records, telephone-based donor mobilisation and rule-of-thumb stock control, which contribute to shortages, expiry wastage and slow emergency response. Artificial intelligence (AI), combined with digital records, Internet of Things (IoT) sensing and blockchain-based traceability, is increasingly proposed to forecast demand, predict donor behaviour and optimise inventory. This paper presents a focused narrative review of peer-reviewed studies on AI in blood management, cov-ering demand forecasting, donor prediction, inventory ordering and traceability. Three connected themes are examined: data sources and sensing technologies, learning methods (time-series models, classical machine learning, gradient boosting and deep learning), and IoT, cloud and blockchain integration. The literature reports that a hybrid forecasting and ordering strategy re-duced hospital red-cell inventory by about 40 % and orders by 60 % against historical practice, that an integrated platform in-creased collected blood by 11 % and reduced wastage by 20 % in a pilot at one centre, and that random forests can identify returning donors with a Matthews correlation coefficient of 0.851. Simple time-series models often remain competitive with deep learning. However, most studies use single-centre or single-country data, retrospective validation and one blood compo-nent at a time, and few report multi-site field trials, complete blood-group coverage or long-term IoT operation. The review contributes a comparative analysis of representative studies, a proposed conceptual layered architecture that links data acquisi-tion, edge and cloud AI, blockchain traceability and independent validation, and a prioritised agenda of research gaps and future directions.
API Security: A Review of Vulnerabilities, Detec-tion Methods, and Emerging Defences
Authors: Prajwal Turankar, Professor Bhosale S.B, Professor Bangar A.P, Professor A.A. Khatri
Abstract: Application Programming Interfaces (APIs) enable communication between applications, services, and distributed systems. As web, mobile, cloud-native, and microservice architectures increasingly depend on APIs, their endpoints also provide opportuni-ties for unauthorized access, data exposure, service disruption, and misuse of security mechanisms. This paper presents a re-view of four recent peer-reviewed sources on API security (a systematic review of security-API misuse detection, a systematic review of web API vulnerabilities and developer challenges, an overview of API protection across the software development life cycle, and a discussion of emerging technologies) and uses their combined findings to build a case for, and a design of, an integrated two-module security framework: a design-time control module aligned with the SDLC, and a runtime protection and monitoring module. The review covers recurring vulnerabilities (broken object-level authorization, weak authentication, exces-sive data exposure, injection, misconfiguration, inadequate rate limiting, poor asset management, and insufficient logging), the human and technical causes of insecure implementation, static, dynamic, heuristic, and machine-learning-based detection ap-proaches, and emerging directions including AI/ML, zero trust, and adaptive authentication. The paper then details a proposed framework architecture, data and tooling requirements, an evaluation plan with specific metrics, expected outcomes, and a dis-cussion of limitations and future work. The overall conclusion is that API security should be treated as a continuous lifecycle responsibility, and that emerging technologies should complement, not replace, fundamental controls.
AI-Driven Unified Data Platform for Oceano-graphic, Fisheries, and Molecular Biodiversity Insights: A Review
Authors: Rupnar Dhanshri Raju, Professor Bangar Abhishek Popat, Professor Bhosale.Sachin Bajirao, Professor Karad Vandana
Abstract: Marine ecosystems produce heterogeneous data from satellites, buoys, research vessels, fisheries activity, laboratory sequenc-ing, and environmental DNA (eDNA) sampling. These sources can reveal ocean conditions, fish distribution, ecosystem change, and biodiversity, but are often stored in separate systems with incompatible formats and metadata. Artificial intelligence (AI) can support forecasting, anomaly detection, habitat modelling, fisheries analysis, and molecular classification. This review examines deep learning for marine water-quality assessment, wireless sensor networks, hybrid models for aquaculture, and molecular biodiversity workflows. It proposes a conceptual unified platform with secure ingestion, source-aware storage, metadata harmonisation, data fusion, AI analytics, and decision-support dashboards. Major challenges include uneven sam-pling, data interoperability, sensor drift, uncertain labels, model transferability, computational cost, and governance of sensitive fisheries and biodiversity data. The platform is a review-derived architecture rather than an implemented or experimentally validated system.
AI-Driven Chatbots in Public Health: A Review of Applications, Evidence, and Challenges
Authors: Satvik V. Sumbare, Professor Bangar A. P, Professor Bhosale S.B, Dr. A. A. Khatri
Abstract: Chatbots driven by artificial intelligence (AI) are increasingly used to deliver health information, screening, counselling, and behavioural support at population scale. This paper reviews five recent secondary studies, namely an umbrella review of 44 systematic reviews, a systematic review of large language model (LLM) chatbots, and three scoping reviews of COVID-19 and public health chatbots, to consolidate what is known about how these tools are designed, where they are applied, and how well they work. Most deployed chatbots are text-based, rely on simple decision-tree or hybrid architectures, and are concentrated in pandemic response, mental health, and health-information provision. Evidence of effectiveness is encouraging for smoking cessation and for short-term reductions in depression and anxiety, but mixed for lifestyle change and clinical decision support, and benefits often fade at follow-up. Persistent weaknesses include a shortage of randomised trials, limited use of theory and user-centred design, inconsistent reporting, readability and accuracy problems in LLM output, slow safety escalation, privacy risks, and digital-divide inequities. The paper closes with priorities for standardised evaluation, domain-specific model devel-opment, safeguarding, and inclusive design.
AI and IoT-Enabled Smart Helmets for Accident Detection, Health Monitoring, and Automatic Emergency Alert Systems
Authors: Saurav S Borase, Professor Abhishek Popat Bangar, Professor S. B. Bhosale, Dr. A. A. Khatri
Abstract: Motorcycle accidents remain a leading cause of death on the roads, particularly in developing countries where two-wheelers are the primary means of transportation. The delayed emergency medical response and lack of information about the victim's health status increases the risk of death during the crucial first hour following the incident. WHO estimates that Powered two-wheelers (PTWs) account for nearly 25% of global road deaths, with some Asian countries witnessing even higher rates due to intense traffic and insufficient protective infrastructure. Researchers have proposed various Internet of Things (IoT) and sensor enabled smart helmets to address these issues. Previous works use a combination of motion sensor and communication mod-ules coupled with micro-controllers to identify accidents and relay user location information. However, a deeper inspection of the recent literature suggests pertinent limitations with the proposed solutions. While most works demonstrate successful crash identification using accelerometers and force sensors, there was an apparent lack of integration of biometric monitoring (heart rate and SpOn), intelligent AI driven severity assessment tool, cloud assisted medical record retrieval and robust false alarm mitigation mechanisms. This paper presents a detailed analysis of five IEEE research papers published in 2023-2025 that ex-plore smart helmet design. The papers examined various existing works related to smart helmet based accident detection sys-tems, user health monitoring and emergency alert relaying mechanisms. The review highlights potential research opportunities based on the gaps identified in the surveyed works. Leveraging the lessons learned from the review and the opportunities high-lighted, this work also proposes an original AI driven smart helmet system design. The proposed model uses a combination of sensors for accident detection, biometric health monitoring and alcohol content estimation. The model also implements an AI based impact assessment algorithm and intelligent dual mode emergency message relaying system. The observations made in this review can form a robust foundation for the development of next generation smart helmet systems that canreduce post-crash mortality rates due to timely emergency response
Digital Forensics Using Memory and File-System Analysis: A Critical Review of Acquisition, Arte-fact Analysis and Correlation Techniques, with a Conceptual Integrated Architecture
Authors: Waghole Krushna Subhash, Professor Bangar Abhishek Popat, Professor S. B. Bhosale, Professor Karad V. A.
Abstract: Evidence on a computer lives in two broad places. Volatile memory (RAM) captures what the system is doing right now, while persistent storage, through its file system, keeps a long-running record of what was created, edited, moved and erased. RAM can hold running processes, open network connections, loaded modules, injected code, credentials and encryption keys, much of which never touches the disk. File systems, in turn, hold files, metadata, journals, timestamps and traces of deleted content. This paper is a structured narrative review of 26 sources (standards, books and peer-reviewed studies) grouped under six themes: forensic principles and process models; memory acquisition; memory analysis and artefact recovery; file-system analy-sis, carving and timelines; solid-state storage and anti-forensics; and scale and tool validation. The literature shows that memory analysis can uncover hidden processes, injected code and key material that a disk examination would miss, whereas file-system analysis rebuilds history and recovers deleted content that memory cannot offer. It also shows that acquisition itself introduces error, because capturing memory changes the machine being captured. Yet the two sources are mostly studied apart, acquisition tools can be evaded or may yield inconsistent images, SSD wear-levelling and TRIM weaken the assumptions behind classical recovery, and growing data volumes are overwhelming manual workflows. The review offers a thematic synthesis, a compari-son of twelve representative studies, a conceptual seven-layer architecture that correlates memory and file-system evidence within an integrity-preserving workflow, and a prioritised research agenda.
Microwave Sensor-Based Systems for Real-Time Detection of Microplastics in Water: A Review of Resonator Sensing, Microfluidic Integration, Machine Learning and Portable System Design
Authors: Prajyot Prakash Vishwasrao, Professor S. B. Bhosale
Abstract: Microplastics (<5 mm) are now found in oceans, rivers, drinking water and human blood, yet routine monitoring is limited because reference methods such as Fourier-transform infrared (FTIR) and Raman spectroscopy are laboratory-bound, slow and costly. Microwave resonator sensing offers a low-cost, non-optical alternative: the large permittivity contrast between water (~80) and common polymers (~2–3.5) means suspended particles shift a planar resonator's resonance frequency, quality factor and transmission/reflection response. This narrative review covers sensing theory, resonator design, microfluidic sample han-dling, machine-learning interpretation (K-nearest neighbours, convolutional neural networks) and portable, wireless integration. Reviewed studies report resonance shifts of ~100 to >500 MHz, a theoretical resolution of ~100 ppm, a limit of detection of 48.72 ng/mL (microfluidic-CNN seawater system), response times below 250 ms, and prediction accuracies of 96–98 % within ±0.01 mg/mL. However, most studies use few polymers, filtered or artificial matrices, laboratory network analysers, non-standard concentration units and random data splits, and few address temperature or salinity compensation or long-term field operation; the seawater study's quoted detection limit and tolerance also need independent confirmation. We contribute a com-parative analysis of representative studies, a conceptual architecture for field-deployable, IoT-connected monitoring, and a prior-itised research agenda.
