Browse Items (16488 total)
Sort by:
-
Comparative Study of AI Models for Automated Tuberculosis Detection Using Image Processing Techniques
Tuberculosis is a critical global health issue, particularly in resource-limited regions where early and accurate diagnosis is important and is in need so that the treatment is effective and the control transmission is controlled. The known diagnostic methods, such as sputum smear microscopy and nucleic acid amplification tests are costly, time-consuming, and require trained professionals. Due this in some cases it is inaccessible in many regions. Deep learning-based automated TB detection offers a promising alternative by enhancing diagnostic efficiency through medical imaging analysis. This study presents a comparative evaluation of five deep learning models, InceptionResNetV2, DenseNet, VGG16, ANN, and a custom CNN, trained on a dataset of 3,008 chest radiograph images, evenly distributed between TB-positive and normal cases. The dataset underwent advanced preprocessing techniques, pixel normalization, and data augmentation. The hyperparameter tuning process was applied, which optimized the learning rates, dropout rates, convolutional filter sizes, and batch sizes to enhance model performance. The models were assessed using accuracy, precision, recall, F1-score, sensitivity, specificity.. Experimental results indicated that the custom CNN achieved the highest classification accuracy (99.51). The superior performance of the custom CNN over other models is attributed to optimized feature extraction, effective preprocessing, and structured hyperparameter tuning. A comparative analysis with previous studies highlights how this approach mitigates dataset limitations and improves model interpretability, and the potential of AI-driven TB detection, enhancing future diagnostic efficiency by improving model generalizability and deployment in real-world healthcare settings. 2025 IEEE. -
A Software Package for Detecting Anomalies in User Authentication
Anomaly detection is a very important tool for various applications such as intrusion detection, fraud, malfunction, system health monitoring and event detection in IoT devices. Recently, user authentication has become an extremely popular topic in information security research environments. The definition of user authentication is formulated as the process of verifying the identity declared by the user for a system object. Authentication is a method used to distinguish between true or false authentication requests. There are many methods used to authenticate a user that can identify valid users in protected resources. This article discusses various methods for analyzing abnormal user behavior in information systems, namely such methods as machine learning, neural networks, hybrid methods. Based on the analysis of system logs in the Astra Linux operating system, a software package has been developed to identify anomalies when trying to authenticate users. 2025 IEEE. -
Using Fog Computing to Accelerate Metagenomic Data Analysis
This article discusses the challenges of processing and analyzing metagenomic data, the volume of which is continuously increasing due to the development of sequencing technologies. Traditional methods such as cloud computing and supercomputing face limitations such as high latency, network dependency, high costs and data security risks. Alternatively, fog computing and hybrid architectures are proposed to distribute the computational load between local devices and cloud systems. This reduces latency, optimizes costs and improves data security. The paper analyzes the advantages of fog computing in metagenomic data analysis, compares it with traditional methods and suggests ways to implement this technology in bioinformatics. The results show that fog computing systems and hybrid systems are promising solutions for applications requiring fast analysis and high data security, such as medical diagnostics and environmental monitoring. The complexity of integrating and managing distributed systems 2025 IEEE. -
Statistical Forecasting of Fat in Body Proportion Utilizing Nonlinear Anthropological Parameters and Density Evaluation
Body Fat Percentage (BFP) is an accurate body fat assessment, plays vital role in order to evaluate an individual's health status and disease risk. Traditional BFP assessments, such as dual-energy X-ray absorptiometry (DXA) and hydrostatic weighing are high in accuracy which is compromised by their cost and complexity. This research work focuses on creating a predictive BFP model using anthropometric techniques. For formulating and validating the proposed model, a benchmark dataset is used consisting of 252 samples having measures of weight, height, waist circumference (WC), hip circumference (HC), skinfold thicknesses along with air displacement plethysmography (ADP) based density estimates. For feature engineering, the most important values are selected such as body mass index, hip ratio etc., as well as logarithmic values and then the best artificial neural network model is trained. The proposed model is developed using quadratic polynomial terms with a literature-based space-cost function (r > 0.98), provided the best model with a Mean Absolute Error (MAE) of 1.5% and coefficient of determination R = 0.92 outperforming conventional works. 2025 IEEE. -
Enhancing Stock Market Forecasting with LLMs, Sentiment Analysis and Technical Indicators
Forecasting stock market trends remains a complex and demanding endeavor due to the intricate and dynamic nature of financial markets. This study explores the combination of sentiment analysis with technical indicators to improve the accuracy of stock price predictions. The research incorporates stock market and news data spanning from January 2015 to June 2023, ensuring a well-aligned and comprehensive dataset. Data was sourced using the Google News API and the Bombay Stock Exchange (BSE), followed by rigorous preprocessing, which involved handling missing values and standardizing sentiment scores for accuracy and consistency. To analyze sentiment, tools like VADER, TextBlob, and the Gemini-1.5-Flash API were employed, with sentiment scores aggregated at the stock level. Simultaneously, key technical indicators including the Relative Strength Index (RSI), Moving Average Convergence Divergence (MACD), Bollinger Bands, and Exponential Moving Averages were derived from stock price patterns. These diverse data points were integrated to predict 14-day closing stock prices, leveraging the Gemini1.5-Flash model for forecasting. The model s performance was assessed using various error metrics such as Mean Squared Error (MSE), Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), and Symmetric Mean Absolute Percentage Error (SMAPE). The results indicated strong predictive accuracy for stable stocks while pointing out challenges in forecasting highly volatile stocks. Ultimately, the findings suggest that combining sentiment analysis with technical indicators strengthens stock market trend predictions, offering a solid foundation for future advancements in real-time financial analytics. 2025 IEEE. -
Advanced Botnet Detection Using Hybrid Machine Learning Models
The improvement of computer network systems, cyberattacks that take advantage of system flaws have increased, resulting in significant monetary losses, business interruptions, harm to one's reputation, and legal repercussions. This research examines nine attack types, those are Fuzzers, Shellcode, Generic, Worms, Analysis, Normal, DoS, Exploits, Backdoor, and Reconnaissance. Botnet assaults are attacks in which a single operator controls several networked devices. The research study examines several models, such as Random Forest, XGBoost Classifier, Logistic Regression, and Decision Tree, to improve detecting skills. By utilizing the advantages of both methods, the suggested ERFwXGBoost (Enhanced Random Forest with XGBoost) model, which blends Random Forest and XGBoost, exhibits remarkable performance. Notably, first they analyze the accuracy, then precision value is also measured, third will measure recall value, and then finally F1 score of the ERFwXGBoost model are all impressively 0.98. In addition to outperforming individual models, our hybrid technique offers a reliable and effective way to detect different kinds of botnet attacks. The research emphasizes how well these models work together to boost overall system security against advanced cyber threats and greatly increase detection accuracy. 2025 IEEE. -
Comparative Analysis of Neural Network Models for Indian Sign Language Hand Gesture Recognition
The recognition of sign language is a crucial element in filling communication gaps that exist in the population. As inclusive communication technologies become more popular, there has been a significant push to develop trustworthy systems for translating sign language into written or visual form. The use of hand gestures and body movements is a fundamental aspect of sign languages, which are commonly used by those who are deaf. The lack of proficiency in sign language makes communication difficult for most people. A project was undertaken to convert Indian Sign Language (ISL) into spoken language through research. The paper presents a comparison of various neural network models. Using OpenCVgenerated real-time images and MediaPipe, it is possible to identify hand movements and collect ISL gesture data in realtime. In the study, it was demonstrated that ResNet50 is 92 per cent accurate in real-time recognition when compared to other models. This work aims to promote inclusivity and communication skills among people who may not have the ability to hear or speak fluently. Adding face recognition to future work may improve accuracy and enable continuous sign language recognition, providing more dynamic and real-time translation capabilities. 2025 IEEE. -
Ensemble Deep Learning for COVID-19 Detection Using Multi-Modal Medical Imaging
The COVID-19 pandemic has had a profound impact worldwide This work proposes a deep ensemble learning model incorporating multi-modal inputs, i.e., CT scans and Xrays, to classify the cases into COVID-19, Viral Pneumonia, or Normal. Employing an ensemble average voting approach from three different CNN models InceptionV3, DenseNet-169, and Xception the suggested methodology is highly accurate and reliable. Preprocessing methods such as Contrast Limited Adaptive Histogram Equalization (CLAHE) improve data quality, and Local Interpretable Model-Agnostic Explanations (LIME) allow interpretable prediction through identification of major image features driving classifications. The ensemble model suggested attains an accuracy of 99.64%, outperforming single models, with precision at 99.50%, recall at 99.73%, and an F1-score of 99.61%, which makes it very reliable for detecting COVID-19. Comparative analysis shows that our ensemble method performs better than individual CNN architectures, such as Xception (99.18%), ResNet101 (98.95%), and DenseNet201 (98.83%), which showcases its better diagnostic performance. 2025 IEEE. -
Federated and Explainable AI Models for Secure FinTech Transactions in Digital Manufacturing Supply Chains
Digital manufacturing supply chains are becoming increasingly dependent on inbuilt FinTech services to perform automated payments, invoicing, and settlements which presents sensitive financial and operational data to security and privacy threats. This article is an empirical paper concerning the application of Federated Learning (FL) and Explainable Artificial Intelligence (XAI) in securing FinTech transactions in decentralized manufacturing supply chains. The suggested framework will facilitate joint fraud and anomaly-related detection without exchanging raw data between supply-chain participants. Different privacy mechanisms such as client-level and secure aggregation are integrated to safeguard sensitive data and minimize the risks of inferences. Explainable AI methods are used such as SHAP, local surrogate models, to enable transparency and auditability as well as regulatory compliance. Experimental evidence has shown that federated models can attain almost centralized detection accuracy with much stronger privacy guarantees and explainability procedures can give insightful and interpretable information about model decisions. The paper identifies the trade-offs between accuracy, privacy, and computational overhead and concludes that federated and explainable AI provides a convenient, secure, and compliant solution to FinTech-enabled digital manufacturing ecosystems. 2026 IEEE. -
AI-Augmented FinTech Platforms for Real-Time Credit Risk and Supply Chain Financing in Smart Industries
A combination of Artificial Intelligence (AI) and FinTech platforms has transformed the financial services sector, specifically, real-time credit risk evaluation and supply chain financing of smart industries. The conventional models of credit assessment, based on the use of fixed financial information and manual processing, are ineffective in capturing the dynamic nature of the modern-day industrial process. This paper conducts empirical research on AI-enhanced FinTech applications and utilizes machine learning (ML), natural language processing (NLP), and multi-modal industrial data, such as financial data, IoT sensor data, and supply chain data, to enable better predictive models and decision processes. The study compares several models, such as ensemble learning and deep neural networks, to predict credit risk and maximize the financing. Findings show that this is highly improved with AUC scores more than 0.88 and reduction in decision latency up to 70 percent showing quicker more information-oriented and context-sensitive risk management. It offers practical implications in the design of AI-based financial solutions, which will allow making smarter credit decisions and allocating working capital in intelligent industries more effectively, and it also notes that AI can transform industrial FinTech ecosystems. 2026 IEEE. -
Edge Attention Module for Object Classification
A novel edge attention-based Convolutional Neural Network (CNN) is proposed in this research for object classification task. With the advent of advanced computing technology, CNN models have achieved to remarkable success, particularly in computer vision applications. Nevertheless, the efficacy of the conventional CNN is often hindered due to class imbalance and inter-class similarity problems, which are particularly prominent in the computer vision field. In this research, we introduce for the first time an Edge Attention Module (EAM) consisting of a Max-Min pooling layer, followed by convolutional layers. This Max-Min pooling is entirely a novel pooling technique, specifically designed to capture only the edge information that is crucial for any object classification task. Therefore, by integrating this novel pooling technique into the attention module, the CNN network inherently prioritizes on essential edge features, thereby boosting the accuracy and F1-score of the model significantly. We have implemented our proposed EAM or 2EAMs on several standard pre-trained CNN models for Caltech-101, Caltech-256, CIFAR-100 and Tiny ImageNet-200 datasets. The extensive experiments reveal that our proposed framework (that is, EAM with CNN and 2EAMs with CNN), outperforms all pre-trained CNN models as well as recent trend models Pooling-based Vision Transformer (PiT), Convolutional Block Attention Module (CBAM), and ConvNext, by substantial margins. We have achieved the accuracy of 95.5% and 86% by the proposed framework on Caltech-101 and Caltech-256 datasets, respectively. So far, this is the best results on these datasets, to the best of our knowledge. All the codes along with graphs, and their classification reports are shared on an anonymous GitHub link: https://anonymous.4open.science/r/Object-Classification-7BE5. 2025 IEEE. -
Transfer Learning based Analysis of Chest X-rays for Accurate Lung Disease Detection and Interpretation
This is a research paper based on a transfer learning approach with a primary aim at the analysis of chest Xrays for accurate detection and interpretation of lung diseases. The proposed method relies heavily on the use of pretrained deep learning models to enhance diagnostic accuracy and reduce the time and computational resources taken during training. Applying transfer learning to a large chest X-ray dataset, the model successfully detects key patterns associated with common lung diseases, such as pneumonia and tuberculosis. The manuscript encompasses data preprocessing, model finetuning, and performance evaluation and demonstrates huge improvements over the traditional methods both in terms of accuracy and interpretability. It has been experimentally proven that the model is competent enough to provide localization of disease areas, as it can be visualized through heatmaps obtained from predictions, which might further help the radiologists perform their diagnosis tasks. This work advocates for medical imaging automation for the early and efficient detection of lung disease. 2025 IEEE. -
Algorithmic Trading and Machine Learning: An Empirical Study of Stock Price Prediction in India
This research paper uses historical data from Ambuja Cement to compare nine machine learning algorithms for algorithmic trading in the Indian stock market. The algorithms applied include SVM, Linear Regression, Decision Tree, K-NN, Ridge Regression, Lasso Regression, Bayesian Ridge Regression, Random Forest, Elastic Net Regression, XGBoost, and reinforcement learning. MSE, MAE, and R2 are metrics used to evaluate predictive performance. The findings show that ensemble approaches and regularized regressions outperform simpler models, emphasizing the importance of model complexity and feature selection. Reinforcement Learning has the potential for optimizing tactics through constant adaptation. The study provides valuable insights on enhancing algorithmic trading in emerging Indian markets. 2025 IEEE. -
SharePort: A Cost Saving and Energy Efficient Ride-Sharing Application for Airport Commutes
SharePort is a mobile application specifically designed for finding people to share rides with, from airport locations. It allows users to find other users at the airport who are traveling to similar locations within a threshold of 2 K M s. This paper talks about the implementation of such an application, its benefit to the society in terms of individual costs, energy savings, traffic reduction, etc. It lays out the design patterns, features and their contribution to the overall idea. In addition to its various positive impacts on the individual expenditure of commuters and the environmental benefits, SharePort also resolves the issue of needing a third-party application to contact the people they are willing to share rides with, by integrating a chat feature, which enables ease of communication and accountability. Initial evaluations demonstrate that pairing up commuters can reduce ride costs by 30-50% per user depending on distance traveled. Additionally, a shared trip also decreases the number of vehicles used for overlapping commutes, reducing fuel burn and carbon emissions compared to independent commutes. Not only does it reduce individual trip costs but it also contributes to lower fuel usage, fewer vehicles on the road, and a more sustainable mobility ecosystem that mutually benefits commuters, cities, and the environment. Relevant SDGs: SDG 11 (Sustainable Cities and Communities), SDG 13 (Climate Action). 2026 IEEE. -
Hierarchical Retrieval Augmentation Generation for Multimodalized Woman's Companion
Empowering Women in society currently face many health-related problems due to the lack of health literacy. Specifically, people are not open to talking about such as sexually transmitted diseases and mental health problems, and counselling is considered taboo in most parts of the world. Some female children grow up under the care of single fathers who are sometimes unaware of the menstrual cycle and the necessary precautions. The solution presented by the research to overcome the problem is a women's health chatbot using Large Language Models (LLM). The research proposes an enhanced retrieval augmentation generation (RAG) architecture that uses the Cloudbased API to get a faster response from the LLM. The women's health chatbot secures data privacy by not saving conversations and being available for 24 hours. Addressing various women's health concerns-such as menstrual health, mental health, pregnancy, and menopause-the chatbot employs the LangChain framework for processing and indexing health-related documents into a vector store for efficient retrieval. The chatbot also features an alert mechanism to identify critical conversations, such as those involving suicidal thoughts, and sends alerts to specified contacts. This integrated approach aims to improve access to accurate health information and support women to make informed health decisions. 2025 IEEE. -
A Cost-Effective NFC-Based Tap-and-Pay Payment System
Rapidly emerging contactless payment methods have totally transformed the economic landscape by ensuring a convenient and efficient means to standardize price structures. Presented in this work is a Tap-and-Pay Payment System which is based on Near Field Communication (NFC) that would substitute the traditional Point-of-Sale (POS) terminals. This device utilizes NFC technology to allow merchants to take payments with only a smartphone, eliminating costly hardware. We provide a comprehensive discussion on the system's architecture, its security properties, the implementation issues faced, and performance evaluations thereof, as well as an investigation into serving SMEs by enhancing the access of digital payments and propagating inclusive finance. Future enhancements include enriching functionalities such as multi-currency support and offline transaction capabilities. 2025 IEEE. -
Precision Farming on Sugarcane: Drone-Based Disease Detection Using YOLOv8 Neural Models
Precision agriculture is being revolutionized by the use of UAVs and AI, enabling more efficient and sustainable crop monitoring. This study presents a drone-based solution for real-time detection of sugarcane diseases such as Rust, Red Rot, Mosaic, and Yellow Leaf. A custom quadcopter, outfitted with a high-resolution camera and Raspberry Pi 4, is used to capture aerial imagery. The onboard YOLOv8 model processes images in real time, with data stored locally on an SD card for further evaluation. The paper covers the complete system setup, including hardware components, neural network deployment, and the end-to-end workflowfrom image capture to decision support. This integrated approach supports early intervention, better yield outcomes, and cost-effective disease management in sugarcane farming. 2025 IEEE. -
Design and Implementation of a Hybrid Solar-Grid Charging Infrastructure with IoT-Based Control
The rapid growth of e-bikes and e-scooters is straining conventional, fossil-fuel-intensive power grids and exposing a critical gap in urban charging infrastructure. In this work, a fully modular hybrid station that synergistically couples three energy vectorsphotovoltaics, second-life Li-ion battery packs, and the utility gridvia an intelligent, sub-50 ms source-arbitration network. The power-conditioning front end employs a flyback-derived SMPS delivering five tightly regulated outputs (5 V, 12 V, 37 V, 48 V DC; 230 V AC) at 9295% efficiency, while a bidirectional synchronous boost stage attains 9497% efficiency and future-proofs the system for vehicle-to-grid operation. End-to-end power quality is preserved with < 5% total harmonic distortion under dynamic loads. An ESP32-centric IoT stack with LoRaWAN back-haul furnishes kilometre-scale telemetry, secure billing, and over-the-air firmware updates, whereas a BiLSTM-assisted battery-management layer enables real-time state-of-charge and state-of-health tracking of repurposed EV modulesextending their usable life and anchoring circular-economy objectives. By fusing high-efficiency power conversion, adaptive energy-source orchestration, and cloud-native intelligence in a compact footprint, the proposed platform sets a scalable blueprint for low-carbon, resilient charging ecosystems that can keep pace with the next wave of urban micro-mobility. 2025 IEEE. -
Temperature and Performance Variations of a Li-ion Battery Pack Under Dynamic Testing Conditions
This study investigates the impact of dynamic mechanical vibrations on the temperature and performance of a 10 Ah, 37 V battery pack. The battery pack was subjected to various mechanical loads using an electrodynamic shaker, while its performance was monitored under different load conditions imposed on a 37 V, 250 W brushless direct current (BLDC) motor. The mechanical load was adjusted using a belt and spring balance arrangement. The batterys parameters, including temperature and performance metrics, were remotely monitored using a combination of sensors, an ESP32 microcontroller, and the ThingSpeak IoT platform. The results indicate that the temperature of the battery pack increased during the shaking process, with corresponding changes in performance. Future research may focus on optimizing battery pack designs to mitigate the effects of mechanical vibrations and improve overall performance under dynamic conditions. 2025 IEEE. -
End-to-End Imitation Learning for Autonomous Driving: Design and Implementation on a Custom Robotic Platform
This paper presents an end-to-end imitation learning system for autonomous driving on a custom robotic vehicle. A PilotNet-inspired CNN, trained on 20 laps of Udacity Beta simulator data from three cameras, was deployed on an NVIDIA Jetson Xavier for real-time steering. The robot features a 48V, 1.5kW BLDC motor and a precision steering system using dual linear actuators. Manual override is enabled via an RC controller. The integration of deep learning with custom hardware highlights challenges in transferring simulation-trained models to real-world systems. 2025 IEEE.