Browse Items (3095 total)
Sort by:
-
Challenges and Opportunities in Deploying Explainable AI for Financial Risk Assessment
Artificial intelligence (AI) has been used more and more in financial decision-making recently, raising questions about the accountability and transparency of these complex systems. The current study investigates the way Explained Artificial Intelligence (XAI) methods might alleviate these concerns and improve the openness of financial decision-making procedures. Nowadays machine learning algorithms are easier to use than ever before, but creating and deploying systems that facilitate real-world banking services has proved challenging. This is mostly due to the fact that algorithms for machine learning are neither transparent or explainable, two attributes that are essential to creating reliable technology. What sets this study unique is the construction of an explainable artificial intelligence (XAI) model that addresses these accessibility concerns while also serving as an instrument for the establishment of credit risk control policies. This work proposes an explainable artificial intelligence model for financing risk control to measure the risks associated with credit financing via peer-to-peer financing networks. The framework uses Shapley parameters to provide AI forecasts according to significant factors that explain. The Support Vector Machine (SVM) and gradient boosting methods had the greatest accuracy scores, 92.4 and 97.6, accordingly. The accuracy of the model was evaluated on a bigger database, and the findings demonstrated that it regularly achieved high levels of accuracy. The SVM and GBM models achieved accuracies of 94.8 and 97.6, respectively. 2025 IEEE. -
The Impact of AI on High-Frequency Trading: A New Paradigm in Share Market Dynamics
A fresh approach in financial trading has emerged as a result of the significant shifts in the dynamics of global share markets brought about by the incorporation of artificial intelligence (AI) into high-frequency trading (HFT). In order to analyse large datasets in real-time, perform trades in microseconds, and AI-driven Deep learning, machine learning, and natural language processing are all examples of things that HFT uses to help people make the best choices they can in markets that have to change all the time. This change will help people make better decisions, and sellers will be able to respond quickly to changes in the market. Individuals can trade faster, better, and for greater amounts of cash with it than ever before. AI does have some problems when used in HFT, though. It can make the market less stable, lead to legal problems, and make it more diligently to be fair and honest. The amount of money, how well markets work, along with the way risks are handled are all changed by AI. It also discusses about how AI changes HFT. This study talks about the pros and cons of HFT powered by AI. Along with the way shares are sold, it also hints at how it might change future rules. 2025 IEEE. -
Algorithmic Crypto Trading using EMA Strategy
Algorithmic trading has transformed financial markets by enabling data-driven strategies that enhance efficiency and decision-making. This paper presents a web-based crypto currency trading platform that employs the Exponential Moving Average (EMA) strategy for automated trade execution, market trend analysis, and portfolio tracking. The platform integrates key performance metrics, including win rate, average profit per trade, risk-reward ratio, and profit factor to assess trading effectiveness. Notably, EMA-based trading achieves the highest profit factor of 3.5 which outperformed deep learning and manual trading by 9.37% and 133%, respectively. Additionally, EMA exhibits a strong win rate of 60%, compared to 65% for deep learning and 40% for manual trading, while maintaining a balanced risk-reward ratio of 2.2. The system features live data visualization, customizable watchlists, and automated trading workflows, providing traders with actionable insights with minimized human error. Performance evaluation indicates that EMA offers a superior trade-off between profitability and risk management, making it a robust and adaptable solution for navigating cryptocurrency markets. This work bridges the gap between manual trading and advanced algorithmic strategies, delivering a user-friendly and efficient trading framework. 2025 IEEE. -
Hybrid Digital Control and Single-Stage Converter Architecture for IoT and Photovoltaic Grid Applications
This work presents a hybrid power conversion strategy that merges high-efficiency digital DC-DC techniques with a universal single-stage DC-DC/AC converter for renewable and low-power applications. From the first concept, a digitally controlled buck converter employing a two-step digital PWM, Adaptive Window ADC, and Self-Tracking Zero Current Detector ensures ultra-low power consumption, minimized output ripple, and improved efficiency, making it suitable for IoT and energy-constrained devices. From the second, a dual-leg single-stage converter architecture enables seamless photovoltaic integration with both DC and AC grids, reducing redundancy through shared semiconductor structures and optimized protection circuits. By unifying these approaches, the proposed system enhances power density, conversion flexibility, and grid adaptability while ensuring reduced switching losses, minimized EMI, and stable operation under varying loads. Simulation and experimental validation confirm efficiency above 91% for low-power DC applications and up to 97% for PV-based grid operation, demonstrating strong potential in sustainable smart energy systems. 2025 IEEE. -
Non-Invasive Detection of Ovarian Cancer using Biosensors Framework with Machine Learning and Federated Learning Techniques
Ovarian cancer is a leading cause of death worldwide, frequently diagnosed at advanced stages due to the lack of effective early screening methods. This work proposes a non-invasive cancer diagnostics utilizing amperometric electrochemical biosensors in early cancer detection from biological fluids, such as urine-based by combination of specific biomarkers like HE4 and Ca125, which are closely associated with ovarian cancer. This study approach integrates machine learning models to work with biosensor data for cancer classification tasks, and federated learning methods to ensure patient data privacy. The proposed system achieves diagnostic results using a synthetic dataset with over 98% accuracy. This decentralized healthcare solution demonstrates early ovarian cancer detection and improved patient outcomes by combining predictive capability with privacy preservation. 2025 IEEE. -
Non-Invasive Detection of Ovarian Cancer using Biosensors Framework with Machine Learning and Federated Learning Techniques
Ovarian cancer is a leading cause of death worldwide, frequently diagnosed at advanced stages due to the lack of effective early screening methods. This work proposes a non-invasive cancer diagnostics utilizing amperometric electrochemical biosensors in early cancer detection from biological fluids, such as urine-based by combination of specific biomarkers like HE4 and Ca125, which are closely associated with ovarian cancer. This study approach integrates machine learning models to work with biosensor data for cancer classification tasks, and federated learning methods to ensure patient data privacy. The proposed system achieves diagnostic results using a synthetic dataset with over 98% accuracy. This decentralized healthcare solution demonstrates early ovarian cancer detection and improved patient outcomes by combining predictive capability with privacy preservation. 2025 IEEE. -
Energy Efficiency Enhancement in Wind-Powered Pumping with TBRC MPPT Integration
This paper introduces a novel strategy to enhance energy efficiency in a Battery and Wind Energy-based Pumping Scheme (BWEPS) by implementing a Test Bench Rapid Control (TBRC) based Maximum Power Point Tracking (MPPT) system within a LabVIEW SPEEDGOAT environment. Wind Energy Conversion Systems (WECS) are inherently challenged by the stochastic nature of wind, which causes frequent fluctuations in output power and reduces overall efficiency if not properly managed. To address these issues, this study applies TBRC as a real-time control framework, enabling faster response, improved adaptability, and more accurate tracking of the maximum power point under dynamic conditions. The integration of battery storage further contributes to stabilizing system performance by mitigating intermittency and ensuring reliable energy availability for pumping operations. The proposed approach not only develops and validates the TBRC-based MPPT algorithm but also optimizes BWEPS operation and benchmarks it against traditional energy storage and control techniques. Experimental validation through real-time simulation demonstrates significant improvements in energy efficiency, reliability, and operational stability. The outcomes highlight the potential of TBRC-based MPPT control as a promising solution for advancing hybrid renewable energy systems, offering an effective pathway for sustainable and resilient water pumping applications. 2025 IEEE. -
A Hybrid Deep Learning and Ensemble Framework for Real-Time Cyclone Path and Intensity Prediction in Disaster-Prone Regions
Predicting the path and strength of cyclones involves significant issues in meteorology, as mistakes can greatly affect disaster management and evacuation strategies. Current models frequently encounter difficulties in achieving accurate real-time forecasting, particularly in representing complicated spatial-temporal dynamics of cyclones. The proposed study presents an innovative hybrid architecture combining deep learning and ensemble methods, using convolutional layers, LSTM units, and a gradient boosting meta-learner to improve prediction efficiency. The system was trained and verified utilising multi-year cyclone datasets obtained from Kaggle, which included atmospheric and oceanic factors. The model architecture attained exceptional accuracy, with a track error of 28 km, a mean absolute error (MAE) of 3.2 hPa for pressure, 4.5 km/h for wind speed, and a root mean square error (RMSE) of 35.4 km. The suggested approach consistently outperformed baseline models, including ConvLSTM, GRU, and XGBoost, across all critical criteria. The deployment in real-time was enabled by a containerised, low-latency API that can integrate with disaster early warning systems. This research enhances cyclone forecasting by offering a scalable, precise, and operationally feasible solution for disaster-prone areas, demonstrating practical superiority over current methodologies. The results highlight the capability of hybrid AI models to improve the accuracy and dependability of meteorological forecasts. 2025 IEEE. -
Redefining Disease Detection: Innovative Machine Learning and Wearable Sensor Integration
Wearable sensor technology is considered to be one of the fastest growing fields of information and communication technologies and it has revolutionized the healthcare delivery by enabling continuous and real-time physiological monitoring. This research presents a novel approach that allows an early onset disease detection instigated with the prowess of advanced Graph Neural Network (GNNs) matched with the body streams gathered from wearable machines using its implementation technology - Pythonline of programming named Awesome Geometric libraries referred to as Aztec PyTorch. Graph neural networks (GNNs) are especially suitable within the scope of modeling complex relationships among multivariate inputs of the sensors for modeling the temporal and spatial subjacent dependence of the physiological signs with regards to reality. The proposed system analyzes the data acquired from the various wearable sensors such as heart rate, accelerometers and bio sensors, which help in anomaly detection and hence the detection of the patient having cardiovascular, metabolic and neurological diseases. The synergy between innovative deep learning models and sensors as ubiquitous technologies offers great promise to transform the provision of personalised healthcare services and dealing with disease in its early stages. 2025 IEEE. -
AI-Driven Predictive Analytics for Sustainable Restaurant Operations and Waste Minimization
There is mounting pressure in the restaurant business to minimize the waste of their operations and use of resources, and still be able to make a profit. Unreasonable forecasting, over-procurement, and poor management of resources are the key causes of environmental and financial waste. As a potential solution to the issue, it presents an AI-based Predictive Analytics Framework (AID-PAF), which combines both a Temporal-Fusion Neural Architecture (TFNA), which is an asset demand prediction framework, and a waste-conscious linear programming model used to solve an inventory and resources allocation problem. Real restaurant operational datasets were used to test the system in a hybrid AnyLogic-MATLAB simulator. Experimental findings show that the proposed framework achieved 40%, 18%, 15%, and 21% reductions in the food waste, energy use, water use, and costs, respectively, and in addition enhanced the accuracy of the forecast, MAPE of 6.5%, the customer fill-rate 96.2%, and the Sustainability Score 78.7. The results prove that predictive analytics based on AI can greatly contribute to the sustainability, efficiency, and profitability of restaurant operations by making intelligent decisions with the assistance of data. 2025 IEEE. -
AI-Driven Tutorial Code Learning System: Personalized Programming Education Through Adaptive Instruction and Gamification
Existing programming education faces critical challenges, such as lack of personalization, restricted feedback tools, and scalability limitations that hinder efficient learning outcomes. This paper presents an AI-powered tutorial code learning system to transform programming education through personalized and adaptive instruction. The system integrates advanced components, including learner modeling, intelligent content recommendation, error analysis, adaptive evaluation, gamification, learning analytics, integration frameworks, quality assurance, security, and scalability layers. To evaluate the system, the study employs a mixed-methods research approach, incorporating embedded case studies and a randomized controlled trial (RCT). Rigorous data collection methods, system measure validation, undergraduate program participant selection, quality assurance protocols, statistical analysis, and ethical considerations are utilized in this work. The architecture demonstrates potential for scalable and globally accessible programming education, addressing traditional challenges through personalized learning protocols. Unlike traditional platforms offering static content and limited feedback, this AI-powered system acts as a personalized tutor, providing active problem-solving and continuous learner engagement. The adaptive system delivers optimal learning paths based on individual student needs and has the potential to transform programming education delivery and outcomes. 2025 IEEE. -
Multivariate Forecasting of Co2 Emissions Using Hybrid Machine Learning Models Based on Energy Consumption and Renewable Adoption
The study presents a machine learning approach to predict carbon dioxide (CO2) emissions by analysing key factors such as energy consumption, renewable energy adoption, and economic growth (GDP). Traditional forecasting methods struggle to capture the complex and nonlinear patterns of emissions. To overcome the limitations and improve the accuracy, research combines classical statistical models like ARIMA and VAR with advance techniques, including deep learning (LSTM) and ensemble methods (XGBoost, stacking). The models are trained on a global dataset of energy and economic records. The results shows that the hybrid models, particularly the LSTM + XGBoost and stacked approaches, have outperformed the traditional methods by obtaining a lower Root Mean Square Error (RMSE) and a higher coefficient of determination (R2). Apart from advancing environmental data science, the research offers a solid predictive framework to support policy initiatives related to the Sustainable Development Goals, specifically SDG 7 (Affordable and Clean Energy) and SDG 13 (Climate Action). 2025 IEEE. -
Screen Time to Severity: Machine Learning Models for Teen Smartphone Dependency Prediction
This study presents a systematic comparison of fourteen supervised classifiers trained to predict binned smartphone addiction levels (Low/Medium/High) in a cohort of 300 teenagers, using demographic, usage, academic, and health related features. After cleaning and binning the continuous Addiction_Level score into three categories, we encoded all categorical variables and standardized inputs, then stratified into 80 % training and 20 % test splits. Our expanded model suite comprised: Logistic Regression, Gaussian Naive Bayes, K-Nearest Neighbors, Decision Tree, Random Forest, Extra Trees, AdaBoost, Gradient Boosting, XGBoost, LightGBM, CatBoost, Support Vector Machine, and a multilayer perceptron (MLP). Each classifier was evaluated on accuracy, precision, recall, macro-averaged F1-score, and multiclass ROC AUC; confusion-matrix entries were flattened into nine 'Actual_i to Pred_j' columns per model for granular error analysis. Logistic Regression achieved the highest test accuracy (98.83%) , outstanding ROC AUC (0.9982) and perfect precision in discriminating the majority class ('High' addiction), despite modest recall for minority classes. MLP followed (96.33 % accuracy, 0.9878 AUC), indicating that a shallow neural network can capture nonlinear patterns but struggles on underrepresented labels. Gradient Boosting, CatBoost, and LightGBM all exceeded 95% accuracy with strong F1-scores (?0.72-0.73) and AUCs above 0.96, demonstrating the power of tree-based ensembles on mixed data types. Simpler methods (e.g., GaussianNB, KNN, Decision Tree) performed moderately (86-91% accuracy, AUC 0.84-0.98), while AdaBoost lagged (77.5 % accuracy, AUC 0.867), suggesting sensitivity to noisy features. Confusion-matrix summaries revealed that most models rarely misclassify Low-addiction teens, but confusion arises between Medium and High classes important for targeted interventions. 2025 IEEE. -
Smote-Enhanced Machine Learning Approaches to Banking Loan Default Prediction: a Multi-Model Study
Accurate prediction of loan defaults is vital for banking risk management, yet loan dataset suffer severe class imbalance, with charged-off loans representing typically less than 10 % of all cases Models trained on such data often exhibit high overall accuracy but poor recall for defaults, limiting their We utilized a stratified 80 / 20 train-test split on a loan dataset dataset of 209,715 loans and 29 features, standardizing numeric variables and one-hot encoding categoricals. Ten algorithmsincluding Logistic Regression, Decision Tree, Random Forest, XGBoost, LightGBM, CatBoost, SGD, MLP, GaussianNB, and KNN were trained without resampling. To address imbalance, we applied SMOTE to the training set, generating synthetic minority instances via k-nearest neighbor interpolation. Baseline models achieved ? 92 % accuracy but recall for defaults ranged 0.04-0.53, underscoring poor minority detection. SMOTE-augmented models saw recall increases up to +0.52 (e.g., KNN: 0.04 ? 0.56) at the cost of reduced accuracy and slight AUC declines, highlighting a precision-recall trade-off. Our systematic multi-model framework demonstrates that SMOTE-enhanced Logistic Regression and KNN markedly improve default recall, offering banks actionable options to prioritize risk detection, while tree-based ensembles retain high ranking performance for applications emphasizing overall accuracy and ROC AUC. 2025 IEEE. -
A Novel Machine Learning Approach for Tuberculosis Detection using Volatile Organic Compounds
For world health, better TB diagnosis is still absolutely necessary. Using VOC Atlas, this study assesses a few machine learning techniques for categorizing breath samples depending on volatile organic compound (VOC) profiles. We created a machine learning pipeline and tried out four different models: Random Forest, XGBoost, Multi-Layer Perceptron (MLP), and a 1D-Convolutional Neural Network (1D-CNN). There were 1,500 patient profiles in the dataset spanning three groups: healthy people, drug-sensitive TB cases, and multidrug-resistant TB cases. Using VOC biomarker patterns found in VOC Atlas and prior TB research, these profiles were developed. While XGBoost stood out by reaching 100% accuracy, our studies revealed that most models performed rather well. This implies that gradient boosting-based ensemble models can adequately grasp the complex patterns found in breath data. One major caveat is that we have not tested these models on real clinical breath samples to validate them. Testing these models with actual patient samples in clinical settings would be the next reasonable step. All told, this research provides a strong basis for creating non-invasive ways to detect illnesses. 2025 IEEE. -
Exploratory Analysis and Pattern Recognition in Energy Production and Demand: A Data-Driven Approach Using Multi-Source Energy Metrics
Hydropower is a leading renewable energy source due to its high efficiency and low operational costs. However, it still faces significant environmental, operational, and forecasting challenges. This paper explores the use of machine learning (ML) models such as SARIMA, Random Forest (RF), and Neural Basis Expansion Analysis for Time Series (NBEATS) to optimize hydropower operations. By analyzing diverse data sets, including hydrometeorological data, plant operations, sensor inputs, and other energy production and demand metrics such as solar, wind, coal, nuclear, and storage, ML enhances decision-making in areas such as inflow forecasting, predictive maintenance, and environmental sustainability. The paper presents an exploratory analysis of 48 -hour energy production and demand patterns across multiple sources (Hydro, Coal, Solar, Wind, Nuclear, and Storage), offering insights into interdependencies and system behavior. It also reviews current ML applications in hydropower, highlights challenges such as data quality and model interpretability, and discusses emerging technologies such as reinforcement learning, explainable AI (XAI), and digital twins as promising future directions. 2025 IEEE. -
QiMINT: Quantum-Inspired Mobile Intelligence - Advancing Complex Signal Processing with Machine Learning
Integrating mobile intelligence with quantum-inspired machine learning (QiML) opens space for challenging mobile signal processing tasks. By utilizing superposition and entanglement, quantum computing principles, QiML boosts the agility of mobile devices by allowing real-time data processing, pattern recognition, and decision making. This work introduces Quantum-Inspired Mobile Intelligence (QiMINT) a new mobile computing framework that integrates quantum-inspired designs with classical machine learning, which increases mobile devices' accuracy, latency, and energy efficiency. Results show that medical QiML based models exceeded traditional machine learning methods in most key performance indicators. The quantum convolutional neural network (QCNN) achieved an accuracy of 92.3% in contrast with the CNNs' 87.5%, but with a processing time of 80 ms as to 120 ms, energy consumption of 10 mJ in comparison to the CNNs' 15mJ. Also, quantum-inspired random forests lowered processing delay by 40%, sustaining superior accuracy than the classical-base system. These results demonstrate that QiML can effectively balance computation complexity while making it suitable for edge computing, IoT, and mobile intelligence systems. 2025 IEEE. -
From Text to Ticker: A Comprehensive Survey and Methodological Guide to Named Entity Recognition in Finance
The financial industry generates vast volumes of unstructured textual data from sources such as regulatory filings, news articles, social media, and earnings call transcripts. Extracting structured and actionable intelligence from this data remains a significant challenge. Named Entity Recognition (NER) is a fundamental task in natural language processing that supports this process by identifying and categorizing key information within text. However, the linguistic complexity, contextual ambiguity, and domain-specific terminology of financial text require specialized approaches that extend beyond general-purpose NER models. This paper presents a comprehensive survey and methodological guide to Financial Named Entity Recognition (Fin-NER). It begins by introducing the core concepts of NER while highlighting the unique challenges posed by financial text. The paper then reviews the evolution of Fin-NER approaches, ranging from rule-based systems and classical machine learning techniques to modern deep learning architectures. Furthermore, it analyzes the distinction between fine-tuned transformer-based models and general-purpose large language models in the current research. The study also examines commonly used datasets and evaluation metrics for benchmarking Fin-NER systems. Finally, it discusses key findings, existing methodological limitations, and future research directions, including hybrid modeling strategies, cross-lingual datasets, and the development of more reliable and explainable systems. Overall, this work serves both as a scholarly review of the Fin-NER field and as a practical guide for researchers and practitioners seeking to transform unstructured financial text into structured and informative representations. 2026 IEEE. -
Analysing Enhanced EEG Based Brain Computer Interface for Motor Imagery Tasks Using Statistical Analysis
BCIs have emerged as a useful tool for helping people with neurological disorders like epilepsy, ALS, and cerebral palsy, who have severely limited communication, by analyzing EEG signals and turning them into actionable signals. This paper aims to investigate the application of BCIs in analyzing EEG signals and designing them to give meaningful signals to the users. Five healthy, right-handed university students (18-21 years) were selected for the study and were asked to spell the words mentally without vocalizing for four cognitive tasks; forward, stop, left, and right. At 100 Hz, EEG signals were sampled and pre-processed with a notch filter and feature extraction was done using DWT to extract important features and KNN was used for classification. All the subjects showed high accuracy with more than 90% and maximum accuracy of 95.89% was obtained by Subject S3. Standard deviations between 1.32 to 1.41, which indicate low variability in performance among all the subjects. These results showed that the DWT with KNN combination can be used for real-time BCI applications and can offer a reliable communication method for people with motor impairment and disabilities. 2025 IEEE. -
Enhancing Prognostication in Colorectal Cancer with Integrated Machine Learning for Improved Survival Prediction
Machine learning methods are recently used to predict patient survival in colorectal cancer using such models as Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), VGG16, and Support Vector Machines (SVM). Taking advantage of a combination of CT, MRI scan images, and clinical records with drug recommendations, the study also checks to see how these models compare for distinguishing between patients in terms of their illness course-whether they are going to get better or worse over time. The results reveal VGG16 has better accuracy than CNN, RNN and SVM; as the highest-performing model tested, it also demonstrates superior precision, recall and F1-score. The research findings also validate these proposed models as they compare favorably with existing literature. This presents a promising proposition: a new, revolutionary approach to using artificial intelligence to boost prognostic accuracy. 2025 IEEE.