Browse Items (16481 total)
Sort by:
-
Comparative Analysis of Machine Learning Models and Interpolation Techniques for Seasonal Rainfall Prediction in Tamil Nadu
This paper explains the rainfall patterns in the state of Tamil Nadu in October 2024, which is the monsoon season, with respect to the differences between the actual rainfall and what is experienced normally over districts. This study uses machine learning techniques from regression models of Random Forest and Gradient Boosting to anticipate future trends about rainfall based on the precedent data. Evaluation using Performance Metrics. The Proposed models are very well tested in terms of performance metrics like RMSE and R-squared, which gives insight about how accurate the forecasts of their results are. This research shows the applicability of QGIS to achieve geospatial analysis for visualizations of the rain distribution as well as anomalies across districts. The current work depicts the integration of data science methodology with geospatial analysis into the knowledge about climate dynamics in the state of Tamil Nadu. Research will help in deepening the understanding of regional climate impacts by bridging predictive analytics with spatial visualization, lending support to informed decision-making in the environment management context. 2025 IEEE. -
Applying a Multi-Agent Simulation Model for Examining Restorative Justice-Based Intervention in the Criminal Justice System: A Legal and Technological Perspective
Previous studies on Restorative Justice (RJ) have focused on the theoretical underpinnings of RJ and its processes. Several systematic literature reviews on RJ point out its potential to assist in victim healing much better compared to the traditional criminal justice system. However, the potential and viability of RJ largely remain in the theoretical landscape. Few empirical studies or simulations have been conducted to explore the viability of this practice in the legal domain. Furthermore, apart from purely studying RJ, literature also points towards its potential use in addressing child sexual abuse cases (CSA) by providing a child and victim-centric approach. However, the practicality of this claim remains scant in present times. Given the gap in the global discourse on the use of RJ as an intervention strategy in the criminal justice system, this paper outlines a computational framework for including RJ into the legal system. The paper does so by applying a Multi-Agent Simulation model (MAS). By utilising JADE for agent orchestration and NetLogo for a visual structure, the framework encodes multiple stakeholders such as accused/ offender, victim, counsellors and judges as autonomous agents with state vectors, utility performance and ACL-communication. The criminal justice system is compared to the restorative justice system using metrics like resolution rate, time, victim healing, rehabilitation and reintegration into the community. Through this paper, a foundation is laid for the potential of RJ in CSA. This paper will enable law and policy makers to consider introducing alternative practices like RJ. 2026 IEEE. -
AI Driven Air Quality Analysis for Health: An Experimental Review
Air pollution, both indoor and outdoor, was linked to 6.7 million premature deaths in 2020, including over 237,000 children under the age of 5, according to WHO. Indoor Air Pollution (IAP) is a crisis of public health that affects billions of people by exposing them to IAP pollutants like particulate matter (PM2.5), volatile organic compounds(VOCs), polycyclic aromatic hydrocarbons (PAHs), and carbon monoxide (CO). The most common cause of IAP varies from incense burning and biomass fuel to ventilation, leading to a horrific human health effect by causing respiratory disease, cardiovascular disease, sick building syndrome, and mental impairment. This review brings together evidence from various studies on the effects of indoor air quality on the environment, health, and productivity. Apart from pollutant exposure, determinants of well-being, i.e., thermal, acoustic, and visual comfort, are the subject of this article. Developments in artificial intelligence (AI), the Internet of Things (IoT), and computational modeling have revolutionized Indoor Air Quality monitoring to detect pollutants and exposures in real-time. All these technologies have the potential to intervene effectively but are intimidating through the prism of high cost, sensor calibration, and the need for large-scale epidemiological studies. To restrict indoor air pollution risks, inter-disciplinary studies need to be adopted to combine effective ventilation technologies and advanced pollutant control systems. Large-scale applications of clean fuel like solar, biogas, electricity, liquefied petroleum gas (LPG), and efficient biomass stoves need to be employed to restrict home air pollution. The present review calls for an emergent public campaign and policy intervention to enhance indoor air quality, health, and well-being. 2025 IEEE. -
A Comprehensive Survey on Decoder Design using Quantum-dot Cellular Automata
QCA offer a compelling alternative to CMOS technology, providing benefits such as low power consumption, high speed, high density, and the ability to surpass the nanoscale limitations of CMOS. QCA is increasingly being adopted in VLSI designs as a solution for reducing power consumption and thermal dissipation. This paper analyzes the area, cell count, and latency of different 2:4 decoders to determine the most efficient design based on these factors. Decoders play a critical role in Quantum-dot Cellular Automata(QCA) by enabling efficient data routing, memory addressing, and logic control while minimizing power consumption and reducing interconnect complexity. The study employs a specialized logic gate known as the Toffoli gate, which is renowned for its capability in reversible computing, allowing information processing without data loss. Future advancements in 2:4 decoders using QCA should prioritize optimizing clocking schemes, improving fault tolerance, and developing scalable architectures to address fabrication challenges and enhance reliability in practical applications. The circuits are simulated using QCA Designer software. 2025 IEEE. -
An Interpretable Federated Multi-Task Learning Framework for Smart Traffic Management with Hessian-Driven Optimization Insights
Smart traffic management faces challenges in balancing privacy, interpretability, and optimization robustness, particularly when using deep learning for vehicle detection and traffic prediction. Existing methods struggle to provide transparent feature attribution while preserving data confidentiality in decentralized settings. This study proposes a federated multi-task learning (FMTL) framework based on YOLOv10, trained on an original traffic dataset, to address these limitations. The framework simultaneously performs vehicle detection, traffic density analysis, and no-entry sign identification, while employing Grad-CAM to enhance interpretability and Hessian-based eigenvalue analysis to evaluate optimization complexity. Results demonstrate an average mean accuracy of 89.7% across three real-world locations, with Grad-CAM revealing meaningful focus on vehicle density and intersection geometry. Hessian analysis confirms the presence of mixed-sign eigenvalues, proving the non-convexity of the optimization surface and highlighting convergence challenges. These outcomes establish a privacypreserving, interpretable, and optimization-aware framework for real-world smart traffic management. 2025 IEEE. -
A Novel Decision Tree and LSTM Powered Intelligent Agent System for Early Detection of Vegetable Plant Diseases
The early and precise diagnosis of vegetable plant diseases is crucial for sustainable agriculture since these diseases have a major impact on crop output and quality. Disease identification performance is examined in this work through a robust detection pipeline that examines the effects of several preprocessing techniques, class imbalance handling strategies, and deep learning classifiers. To better represent data and increase model knowledge of illness characteristics, the GLCM was used to extract texture information. By combining XGBoost with LSTM networks, a new hybrid model was created. While XGBoost is great at classifying structured data, the LSTM component is great at evaluating sequential data, which allows it to capture patterns and trends in the evolution of plant diseases over time. Better and more meaningful forecasts are made possible by this supplementary integration. By surpassing more conventional methods of illness classification, the suggested LSTM-XG model attained a remarkable prediction accuracy of 99.34%. An important factor in achieving this outcome was the use of hybrid modeling in conjunction with thorough preprocessing and correction of class imbalance. Finally, the LSTM-XG model shows great promise for practical use in precision farming. Its precision and efficiency in identifying illnesses in vegetable plants might facilitate prompt action, lessen crop loss, and encourage better farming methods. 2025 IEEE. -
D-GRAM: Dynamic Game-Theoretic Risk Modeling for Adaptive Cyber Defense
Cybersecurity threats are increasing in scale and sophistication, requiring strategic decision-making for cost-effective defense. This work presents a non-cooperative game-theoretic framework to model the interaction between a rational attacker and a defender with limited resources. Each player selects from a finite set of strategies, and payoffs are computed dynamically based on the probability of attack success, defense cost, and potential impact. A loss-risk relationship is used to populate the payoff matrix, ensuring that outcomes reflect realistic operational conditions. A mixed-strategy Nash equilibrium is calculated to determine optimal attack and defense probabilities, thereby balancing resource use and risk mitigation. To improve practicality, an adaptive defense mechanism is introduced, allowing the defender to update strategy probabilities incrementally based on observed attacker behavior. Sensitivity analysis reveals how equilibrium strategies adjust in response to variations in attack cost, defense cost, and impact severity. The results highlight how adaptive learning enhances resilience while minimizing unnecessary defense efforts, making the approach suitable for resource-constrained network environments. 2025 IEEE. -
Artificial Intelligence and Human-AI Driven Accreditation System for Higher Education Quality Assurance
This study introduces the Artificial Intelligence and Human-AI powered Accreditation System, which is meant to transform quality assurance in higher education. The suggested framework will combine human knowledge and smart automation to guarantee transparency, scalability and reliability of accreditation assessments. The data processing algorithm is based on Robust Principal Component Analysis of noise elimination and data normalization and then on the minimum-Redundancy Maximum-Relevancefeature selection algorithm. The fundamental classifier is a Graph Attention Network developed on PyTorch and PyTorch Geometric that is able to capture all the relations between institutional characteristics to make explainable judgments. A blockchain ledger is incorporated to record accreditation results to achieve security and traceability. The experimental simulations show excellent performance with high accuracy, evaluation times, and fairness than the traditional models. The system offers a strong, smart and open accreditation system that facilitates continuous improvement and accountability in the higher education system. 2025 IEEE. -
Object Detection Framework for Identifying Suspicious Items in School Environments using YOLOv8
The issue of unattended bags, metallic items, and concealed weapons at schools has made school safety a growing issue worldwide. This paper presents a software-driven, deep learning framework, implementing automatic identification of suspicious items within a school environment, using the new YOLOv8 neural net architecture. A proprietary 5,000 image dataset of simulated school corridors and classrooms, with 5 annotated threat classes, was developed. The system attained a mAP of 95.6%, precision of 96.8%, and recall of 94.5%, with 38 fps inference speed using a single GPU. YOLOv5 and Faster RCNN comparisons showed a mAP improvement of 12-15%, along with nearly 2x faster frame throughput for the proposed approach YOLOv8, resulting in lower latency and faster responsiveness. The system works in a real-time framework, producing annotated alert logs and frames with mAP scores above 0.5. Experiments conducted with different levels of clutter and illumination show the system has sufficient robustness for school surveillance use cases. 2025 IEEE. -
Attention-based CNN for Adversarial File Fragment Detection Against Padding and Bit-Flip Attacks
File fragment classification represents a critical task within digital forensics and cybersecurity that aims to recover fragmented files when their metadata is not available. Even though cutting-edge deep learning models achieve 77-79% accuracy on clean fragments, none of the existing file fragment classification systems currently include detection mechanisms against adversarial attacks, thus remaining defenseless against attackers using byte-level perturbations. This paper addresses this gap by proposing the first adversarial detection framework for file fragment classification. This paper presents an attention-based CNN that combines byte embeddings with both spatial and channel attention mechanisms to detect byte-level perturbations before actual classification. Evaluated over 30.72 million fragments across 75 file types, the detector reaches an accuracy of 91.44% against five attack strategies: null-byte padding, random-byte padding, cross-file padding, random bit-flipping, and header-targeted bit-flipping, at 91.34% recall, 95.46% specificity, and 0.9819 AUC-ROC. With 1.31 M parameters and 1 ms inference time per fragment, the detector enables practical deployment as a preprocessing filter within two-stage forensic pipelines screening suspicious fragments before reaching standard classifiers. This foundational work sets up the first comprehensive benchmark for adversarial robustness evaluation specifically in file fragment classification. 2025 IEEE. -
Forecasting Global Microplastic Exposure from Processed Foods: Data-Driven Forecasts and Detection
Microplastics are one of the major contaminants of processed foods at a global scale and they contain high risks for human health. Even though the public understanding of the issue has become wider, the knowledge of individual levels of exposure is still very much limited together with the practical tools which can estimate microplastic ingestion. This study proposes a complete data pipeline and a machine learning framework for predicting microplastic contamination and estimating personalised exposure to microplastics depending on country, specific consumption patterns and contamination trends of a long, term nature. The dataset consisted of approximately 18 food groups across 109 countries. So far the data has been through a very thorough preprocessing stage, exploratory analysis, and feature engineering was undertaken, which among other things, included microplastic load aggregation, the addition of lagged variables, and mixing serving sizes information. Random Forest and XGBoost regressors models were trained to predict the levels of contamination from 2019 to 2030. Polynomial Regression delivered the highest accuracy on the training data of R2= 0.9897. While XGBoost gave the best generalization result of R2 = 0.9469 and was therefore chosen as a final forecasting model. The consumption of microplastics through the global food chains is predicted to keep increasing. The originality of this study is in the combination of the long, term contamination data with the selective food, category modelling that allows to generate a reliable framework for the forecasting of the individual intake and to provide to the policy makers EBP (Evidence, Based Policy) advice. 2025 IEEE. -
A Hybrid Genetic Algorithm and Large Language Model Approach for Agricultural Products Price Optimization
This paper introduces a hybrid approach, based on Genetic Algorithm (GA) and Large Language Models (LLMs), namely Mixtral 8x7B, to optimize pricing strategies for agricultural products. The method processes real-time market data, using Machine Learning (ML) techniques to generate competitive and profitable price recommendations. GAs allow for adaptive optimization, while LLMs capture complex trends in the market, making this approach more precise with respect to the pricing strategy. Case studies related to onions and tomatoes illustrate the efficiency of the optimization process. The outcome shows that the optimized prices achieve a fitness score of 0.915 (onions) and a competitive index of 0.89 (onions) compared to the market averages. Compared to traditional methods, the proposed hybrid model provides a better approach towards decision making through multi-objective optimization and real-time data analysis. This research contributes to improved profitability for farmers by adopting sustainable pricing strategies and agricultural market efficiency. 2025 IEEE. -
Enhancing Satellite Imagery with GAN Based Cloud Removal
Satellite imaging is one of the most common uses for applications agricultural, urban planning and environmental monitoring to mention a few. Unfortunately, even the best-laid plans for aerial photography can be decimated by one thing: cloud cover. The novel way of cloud extraction from the satellite data that is demonstrated in this article, use a Generative Adversarial Network (GAN). For the betterment of cloud removal, ResNet based discriminator and a UNet-based generator are utilized in the suggested approach. To accurately train the networks, a new technique was also developed to introduce noise that resembles natural cloud patterns. The PSNR score, as a qualitative and quantitative index card that uses the PyTorch- based GAN methodology to verify different performances in traditional methods based on EuroSat. 2025 IEEE. -
EUI: A Novel Underwater Image Enhancement Network
The use of optical imaging cameras on underwater vehicles has been increasingly common in recent years. These cameras are increasingly being utilized for the purpose of assisting with the search for aquatic items and the gathering of images. The past ten years have seen the publication of a number of techniques that have been utilized to enhance underwater images. These techniques include enhancing the signal-to-noise ratio and decreasing the amount of backscattered noise at the receiving end. Nevertheless, the development of these algorithms was driven by the need to find a solution to the problem of improving photographs of underwater environments when they are exposed to daylight. The accuracy of these, on the other hand, will not be known until they have been tested on photographs taken underwater in low light conditions. As a result of the fact that dark underwater scene photos typically have an exceptionally low quality and the presence of a great deal of noise, it is simple for artifacts to arise during the process of improvement. In order to fill this need, we conduct an in-depth analysis of the most recent deep learning-based algorithms for the enhancement of underwater images. A novel underwater image augmentation network that is capable of handling the severe decrease in underwater image quality that is induced by low illumination is that which we propose as our last suggestion. Our approach allows for the possibility of ULPs avoiding both low-light damage and scattering at the same time. Additionally, the results of our tests suggest that our method continues to be trustworthy even when exposed to different levels of illumination, which has allowed it to be applied to a wider range of applications. When compared to some of the most cutting-edge strategies for improving underwater pictures that are already in use, as well as techniques for improving low-light images, our method has shown to be superior in terms of performance in a variety of low-light underwater situations. 2025 IEEE. -
EUI: A Novel Underwater Image Enhancement Network
The use of optical imaging cameras on underwater vehicles has been increasingly common in recent years. These cameras are increasingly being utilized for the purpose of assisting with the search for aquatic items and the gathering of images. The past ten years have seen the publication of a number of techniques that have been utilized to enhance underwater images. These techniques include enhancing the signal-to-noise ratio and decreasing the amount of backscattered noise at the receiving end. Nevertheless, the development of these algorithms was driven by the need to find a solution to the problem of improving photographs of underwater environments when they are exposed to daylight. The accuracy of these, on the other hand, will not be known until they have been tested on photographs taken underwater in low light conditions. As a result of the fact that dark underwater scene photos typically have an exceptionally low quality and the presence of a great deal of noise, it is simple for artifacts to arise during the process of improvement. In order to fill this need, we conduct an in-depth analysis of the most recent deep learning-based algorithms for the enhancement of underwater images. A novel underwater image augmentation network that is capable of handling the severe decrease in underwater image quality that is induced by low illumination is that which we propose as our last suggestion. Our approach allows for the possibility of ULPs avoiding both low-light damage and scattering at the same time. Additionally, the results of our tests suggest that our method continues to be trustworthy even when exposed to different levels of illumination, which has allowed it to be applied to a wider range of applications. When compared to some of the most cutting-edge strategies for improving underwater pictures that are already in use, as well as techniques for improving low-light images, our method has shown to be superior in terms of performance in a variety of low-light underwater situations. 2025 IEEE. -
Hybrid Model Integrating Firefly Algorithm and Gradient Boosted Trees for Stronger Financial Fraud Detection
Financial fraud poses a critical threat, with complex attack patterns overwhelming traditional rule-based detection methods. This project proposes a hybrid model integrating the Firefly Algorithm (FA) for hyperparameter optimization with Gradient Boosted Trees (GBTs) to enhance fraud detection. Evaluated on an imbalanced financial dataset, the FA-GBT framework achieved significant performance gains. ROC-AUC increased by 5.08% to 0.9883, demonstrating superior discrimination. Critically, Precision rose by 13.38%, substantially reducing false positives/false alarms and operational costs. The optimization also reduced training time by 64%(from 9.8 seconds to 3.5 seconds). This supports a robust, high-confidence, and efficient strategy for deployment in real-time transaction systems. 2025 IEEE. -
Breast Cancer Diagnosis: Feature Selection and Ensemble Machine Learning
Breast cancer diagnosis requires accurate diagnostic tools that are both efficient and interpretable for clinical deployment. This study presents an integrated pipeline combining Recursive Feature Elimination with Cross-Validation (RFECV), Synthetic Minority Over-sampling Technique (SMOTE), and ensemble learning methods applied to the Wisconsin Breast Cancer Diagnostic dataset. RFECV achieved dimensionality reduction from 30 to 17 features, representing a 43% reduction while maintaining predictive performance. SMOTE transformed the class imbalance ratio from 1.68:1 to a perfect 1:1 balance. A comprehensive evaluation of twelve machine learning models revealed that LightGBM attained an F1-score of 0.9722, accuracy of 96.5%, and ROC-AUC of 0.9914 with strong cross-validation stability (0.9681 0.0179). Feature importance analysis identified worst perimeter, area, and concave points as the most discriminative features for differentiating malignant from benign tumors. The proposed approach achieved a 35% reduction in training time compared to full-featured models without sacrificing performance. This reproducible pipeline demonstrates practical clinical relevance for automated breast cancer diagnosis with improved computational efficiency and model interpretability. 2025 IEEE. -
Fine-Tuning Large Language Models for Personality Development
Large Language Models (LLMs) are state-of-the-art in Natural Language Processing (NLP) tasks but are extremely challenging to fine-tune with their size and computational needs. The current research targets the fine-tuning of the OpenLLaMA-3B model for personality development tasks with parameter-efficient fine-tuning (PEFT) via LoRA and QLoRA. To deal with hardware limitations, 4-bit quantization through BitsAndBytes was used, lowering memory without affecting accuracy. A FAISS-based Retrieval-Augmented Generation (RAG) pipeline was also used to improve contextual reliability. Semantic similarity (cosine similarity), BLEURT, ROUGE, and human evaluation were used to evaluate the model. The results of this experiment show that semantic similarity greatly increasing from 0% before fine-tuning to 80% after fine-tuning, demonstrating the benefit of PEFT and quantization in domain adaptation. The work illustrates that by leveraging effective fine-tuning, quantization, and retrieval augmentation, LLMs can be deployed at scale under limited resources while providing high contextual accuracy for personality development purposes. This approach not only enhances knowledge of the job, but it also has practical scalability for educational and self-improvement purposes. The findings highlight a viable path forward for applying small, yet strong, LLMs to personalized learning and adaptive human-AI interaction. 2025 IEEE. -
A Hybrid Ensemble Model Combining Machine Learning and Fuzzy Logic for Robust Stress Level Detection
Accurate stress detection remains challenging because stress manifests differently across diverse populations. This research presents a novel hybrid approach that combines four machine learning classifiers - XGBoost, Random Forest, Support Vector Machine, and Logistic Regression - with an enhanced fuzzy logic system through weighted voting. The study employed a comprehensive dataset comprising over 1,100 samples containing more than 45 features that capture psychological, academic, social, physical, and environmental stress indicators. The results demonstrate that the ensemble achieves 87.27% accuracy with statistically significant improvements over individual models (p0.001). The approach maintains strong performance when 10% noise is injected or missing data is simulated, and provides well-calibrated probability estimates that could support clinical decision-making. 2025 IEEE. -
High-Efficiency Five-Level Multilevel Inverter for Grid-Connected Renewable Energy Systems: Modeling and Simulation
This paper describes the design of a 5-level multilevel inverter for grid connected renewable energy systems simulated using MATLAB. The inverter developed works with an input voltage of 164 V and a well-regulated output voltage of 325.4 V, with the switching frequency fixed at 20 kHz for efficiency, while controlling the modulation index (M) for proper voltage regulation. The performance review of the inverter such as THD and power factor were simulated with MATLAB. It is found that the voltage quality has been greatly improved and harmonic vicious generation has been diminished when compared with traditional inverter. The high-quality of output voltage provided by the inverter also makes it ideal for use with renewable energy systems. The simulations also demonstrate the efficiency of the inverter when controlling power for grid connection, revealing its capability to offer secure and low-cost renewable energy solutions. Performance evaluations indicate improved voltage waveform quality and higher efficiency of the system, which makes it promising for future renewable energy connected applications. The fact of using MATLAB simulations to verify our designs further proves its feasibility and efficacy in handling the challenges of power conversion in sustainable energy systems. 2026 IEEE.