Browse Items (3095 total)
Sort by:
-
Depression Severity Prediction Among Higher Education Students Using Neural Network Model
Depression significantly affects students' mental health and academic performance, highlighting the need for effective early detection methods. This study investigates machine learning approaches for automated classification of depression severity using responses from the Patient Health Questionnaire-9 (PHQ-9). Deep Neural Network (DNN), Long Short-Term Memory (LSTM), and hybrid models combining structured PHQ-9 scores with descriptive text responses were evaluated. The experimental results show that the LSTM model achieved the highest classification precision (90%), demonstrating its ability to capture sequential relationships between items in PHQ-9. The findings indicate that sequence-based models are well suited to assess the severity of depression in student populations. Integrating such predictive models into digital mental health screening systems may support the early identification of at-risk students and enable timely, data-driven interventions in academic settings. 2026 IEEE. -
Alzheimer's Disease Detection using Deep Feature Extraction and Explainable Machine Learning
Alzheimer's disease (AD) is a progressive neurodegenerative disorder characterized by gradual cognitive decline, posing significant diagnostic challenges that necessitate automated detection systems to aid clinical decision-making. This study presents an explainable machine learning framework for binary dementia classification using deep feature extraction from magnetic resonance imaging. A pretrained ResNet50 convolutional neural network was employed to extract 2048-dimensional feature vectors from 86,437 MRI slices derived from the OASIS1 dataset, encompassing 347 subjects. The dataset was imbalanced, containing 67,222 Non-demented and 19,215 demented slices (combining very mild, mild, and moderate dementia). The aggregated features at the Subject-level were used to train three machine learning classifiers: Logistic Regression, Random Forest, and XGBoost. The XGBoost model achieved the highest accuracy of 77.14, with a precision of 0.84 and a recall of 0.87 for Nondemented cases, demonstrating strong discriminative capability. Gradient-weighted Class Activation Mapping (Grad-CAM) visualizations highlighted the hippocampus and temporal lobes as key regions influencing predictions, aligning with established Alzheimer's pathology. The study demonstrates the potential of combining deep feature extraction with interpretable machine learning for automated dementia screening. 2026 IEEE. -
Enhancing Sign Language Recognition Through LSTM Model
Sign language recognition is a remarkable task in this project completed through two state-of-the-art methods, Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU). This way, the system is able to quickly process each frame of the webcam with real-time information regarding face, body and posture in order to extract critical values. this research seeks to provide the necessary resources and opportunities for deaf people to be able to communicate effectively, obtain an education and enjoy their lives as much as other human beings This makes it a very important tool for education where the system can convert sign motions into text on-the-fly. The data was collected through a live camera, and key points from face, body, and pose were detected for training the model. Kindergarten used the four categories of vegetables, fruits, colors and animals. There were 40 video sequences of 40 frames with a sign in each. the model tries to fit too much to noisy points of data. However comprehensive the training, after 19 epochs the validation accuracy is an impressive 93%. The oscillations in the truth values of models are indicative of some uncertainty in learning where the accuracy is still to be settled. The graph in general shows that the LSTM based sign language movement classifier has a good capacity to learn and identify sign language movements with high precision. 2025 IEEE. -
Machine Learning Models for Apple Disease Detection With Texture Feature Fusion and Feature Selection
Computer vision has become an integral part of modern agriculture. One of the key applications of Computer vision is the automatic detection and classification of plant disease from digital images of plant leaves. In this study we evaluate the discriminatory capability of selected texture features and their fusion in identifying plant diseases from leaf images. Further, the performance of four feature selection algorithms is also evaluated. Texture features are extracted from resized raw images. Experiments are carried out with public data sets of Apple plants. Through extensive experimentation, two classifiers - Random forest and XGBoost are chosen for the evaluation. The feature fusion and feature selection resulted in 85% accuracy. The result is promising as the features are extracted from whole leaf images, without any segmentation. 2025 IEEE. -
Predictive Modelling of Microwave Link Failures Using Machine Learning and Deep Learning Techniques
Microwave radio links play a vital role in keeping mobile networks running, especially when it comes to backhaul-the part of the network that connects base stations to the core. Gradual failure in these links could disrupt services and cost providers a lot in both revenue and customer trust. In this study, we explore how machine learning can help predict such failures before they happen. Network performance data from a mobile network operator in Nigeria was collected, cleaned and used to achieve the purpose of the study. Four algorithms belonging to machine learning (ML) and deep learning (DL) were adopted and used for training the dataset and predicting link failures. Results show that the Long ShortTerm Memory (LSTM - a deep learning model effective for handling time-series data) performed best with prediction accuracy of 92%, distantly followed by others. These findings indicate that the LSTM is better in modelling temporal patterns in network behaviours. This study provides a practical framework for automating microwave link monitoring and maintenance, thereby reducing manual diagnostics, preventing outages, and improving service reliability. The proposed solution supports the integration of predictive intelligence into network operations, enhancing the quality of service and operational efficiency for telecom providers. 2025 IEEE. -
Comparative Analysis of Machine Learning Models for Uterine Cancer Prediction Using Clinical and Genomic Data
Uterine cancer prediction accuracy is important in clinical decision-making because it improves the overall chances of patient recovery. Several machine learning models, such as Decision Tree, Random Forest, XGBoost Regressor, and Support Vector Regressor, were explored to determine which is more effective in predicting uterine cancer. Attributes such as mutation counts, diagnosis age, and MSI score, were used for the analysis. The different models were tested using the standard performance metrics such as the Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and R2 Score. Random Forest showed the highest predictive performance with an R2 score of 0.655, followed by XGBoost regressor, which was relatively close to the R2 score of Random Forest. Support Vector Regressor performed very poorly as the R2 score was negative, implying that the model is not suitable for such prediction. Ensemble-based models, which include Random Forest and XGBoost Regressor, have proven to be more effective in handling medical prediction tasks, and this is because of their robustness and their ability when it comes to handle overfitting. Though model generalizability was affected due to small data size and the absence of hyperparameter tuning. The future work will focus on expanding the dataset, implementing hyperparameter tuning, integrating deep learning, and leveraging explainable AI (XAI). The research has provided valuable insight for clinicians who wish to use machine learning for uterine cancer prognosis. 2025 IEEE. -
Harnessing Behavioural Insights for Autism Spectrum Disorder Prediction via Machine Learning
This paper exploration has been done for predicting autism spectrum disorder or ASD while using certain machine learning classifiers. The multisource dataset used, covers behavioral, genetic, neuroimaging and sensor data. The classifiers used are: Random Forest, AdaBoost, Extra Trees, Logistic Regression, K Neighbors, and Bernoulli Naive Bayes. The results achieved are: The Bernoulli Naive Bayes (92.10%), Random Forest (91.53%) and Logistic Regression (91.34%). AdaBoost and Extra Trees performed well with accuracies 90.34% and 90.20%, respectively. K Neighbors Classifier had the lowest accurate outcome with 87.65%. This study explores improving ASD diagnosis, highlighting the effectiveness of various models and emphasizing the need for further research to address challenges such as model interpretability and data quality. 2025 IEEE. -
Detection and Sensing of Human Body Micro-Motions Using 24GHz mm-Waves: A Case Study
The paper presents a 24 GHz millimeter-wave (mm-Wave) system for real-time human presence and distance detection, leveraging the 24 GHz band's balance of range, resolution, and power efficiency for applications in smart environments. The system uses Doppler-based reflections from signals to accurately detect presence and estimate distances with 12cm accuracy up to 5 meters with little needed infrastructure. The challenges presented by signal attenuation and multipath interference are addressed using beamforming and Massive MIMO with 5G-enabled IoT as the framework. The use of Ultra-Reliable Low Latency Communication and Massive-Machine-Type Communication allows a framework for rapid data processing and scalability. This system enables automated smart home lighting, healthcare occupancy detection, and security intrusion alert applications, with very high detection accuracies. The limits are reduced detection performance beyond a distance of 6 meters and interference from reflective surfaces. Future work includes investigating the 60 GHz bands, which will yield higher resolution in the detection, and using machine learning techniques to develop adaptive detection, as a scalable and cost effective method of real-time automation in a range of settings. 2025 IEEE. -
Prediction of Campus Energy Consumption Patterns Using Machine Learning Techniques
The exponential increase in campus energy consumption results from the rise in population density, leading to urbanisation and the use of higher energy-intensive devices within the environment. This study explored high-performance data analytics techniques to visualise energy consumption across buildings using datasets obtained from a load audit of the entire distribution network within the Federal University of Technology, Owerri (FUTO). Advanced time series models were used to predict and forecast the consumption patterns for a year. Visualisations for this research provided detailed insights into the energy profile across all the clusters, while the SARIMA, ARIMA, and Prophet models predicted the energy demands. The heatmap for the correlation matrix reveals a constant energy scale throughout the week (weekend average energy usage is at least 40% of the weekday). A comparative performance was done to analyse the scalability and predictive abilities of the individual models. Results from the study indicate that SARIMA has the lowest mean square error (4.4896) and the highest R2 score (0.8362). The study concludes that the adoption of machine learning models for energy forecasting and prediction is vital for modern-day energy management in the University. 2025 IEEE. -
Deep Learning-Based Health Risk Prediction in Contact Sports Using Wearable Sensor Data
This study presents a deep learning-based approach to predicting physiological health risks in athletes engaged in contact sports using wearable sensor data. Motivated by the need to detect early warning signs of collapse or severe fatigue, this study employs a Long Short-Term Memory (LSTM) neural network to analyse multivariate time-series data. Key physiological signals, including heart rate, body temperature, and motion, were extracted from the PAMAP2 dataset to train and validate the model. The LSTM demonstrated strong predictive performance, achieving an accuracy of 98.3% in identifying potentially dangerous physiological states. In addition to its high classification accuracy, the model effectively captured temporal dependencies in the data, underscoring its suitability for health risk prediction in dynamic, high-intensity sports environments. This study highlights the potential of wearable data and LSTMbased analysis in supporting proactive athlete health management and injury prevention. 2025 IEEE. -
AI in Predictive HR Analytics for Talent Management
This paper presents the topic on how Machine Learning (ML) can be used to conduct Predictive HR Analytics to streamline Talent Management practices. The aim of the development of the project is mainly the application of Random Forest as a supervised learning model to forecast turnover of the employees, performance, and career-growth potential. With the historical employee data, such as performance reviews, tenure, and levels of engagement, Random Forest models would help determine the aspects that are significant factors to employee retention and performance. The model is incorporated with HR software solutions such as SAP SuccessFactors that help to gather information seamlessly to make predictions in real-time, and base decisions on data. It can be seen in the findings of this research that this approach to identifying the factors that influence the effort to retain employees based on the likelihood of them leaving was not only more accurate than other methods but much more effective in the retention efforts. Through predictive analytics, organizations are better placed to take the initiative of managing talent, minimizing turnover and streamline workforce productivity, which eventually lead to business success. This research demonstrates that such predictive models based on AI have a high potential to change HR practice. 2025 IEEE. -
Cognitive IoT-Integrated Real-Time Transaction Monitoring System for Financial Security
As financial transactions become more complex in terms of the digitally connected environment, smart and real-time monitoring systems are required to fight off fraud and guarantee safety. The paper proposes a Cognitive IoT based Real-Time Transaction Monitoring System which is a proposed system to the financial sector. The proposed model resorts to the Z-Score normalization as a means of pre-processing the data and Recursive Feature Elimination with Random Forest as a method of selecting the most relevant features affecting the behavior of transactions. As the main type of classifier, a hybrid LSTM-Autoencoder model is used that makes it possible to reliably detect anomalies both using sequential and reconstruction-based learning. Developed on top of TensorFlow and its federated learning extensions, the system maintains the privacy of its user data by supporting decentralized training of edge devices. Experimental tests indicate higher results over the detection accuracy, which reduces false positives, and real-time reactivity. It is a scalable, safe, flexible system that can be used to monitor financial transactions efficiently, using the combined potential of the cognitive computer, the IoT environment, and powerful machine learning to respond to the dynamic nature of financial security issues. 2025 IEEE. -
Market-Based Strategies for Enhancing Grid Resilience under High Renewable Penetration
The rise in penetration of renewable energy sources (RES), especially wind and solar, presents important challenges to power system operation and electricity markets. Renewables cut emissions and are cheaper in the long run but are intermittent, creating volatility, uncertainty about revenues, and reliability risks. This study analyses the performance of markets in delivering resilience to the grid given high levels of renewables. A grid simulation over 24 hours was then developed in order to study four aspects of the problem; (i) Stability of revenues for the generators, (ii) Reduction in the cost of imbalances as a result of flexibility, (iii) Reliability in terms of Loss of Load Expectation (LOLE), and Energy Not Served (ENS), and (iv) the impact on costs for consumers. Findings show that flexibility participation reduces costs incurred due to imbalance by ? 60%, and resilience mechanisms reduce the effective tariff faced by consumers from 32/kWh to 6.8/kWh. Reliability indices also improve greatly when flexibility and ancillary services markets are introduced. In addition, a Particle Swarm Optimization (PSO) model was applied to find the optimum levels of renewable penetration and flexibility that result in the least total system cost, which correspond to approximately 0.7 for flexibility and approximately 55-65% for renewables. These results point to the economic and technical need for reengineered electricity markets that incorporate flexibility products, ancillary services, and resilience incentives. This dissertation provides a complete methodology of simulation, reliability metrics, and optimization to inform policy makers, system operators, and investors of resilient renewable dominated grids. 2025 IEEE. -
Counterfactual Demand Forecasting Using Multivariate LSTM
Demand forecasting is a key part of running operations efficiently in the fast-changing retail and online shopping industries. Regular methods that use statistics often have trouble handling the complex, changing, and time-based patterns found in actual sales data. This study introduces a new way to predict demand that uses multivariate Long Short-Term Memory (LSTM) models. The models take both the order of sales over time and other factors like prices and weather into account. Three model designs were tested: a simple straightforward model, a pure LSTM model, and a new hybrid LSTM model that mixes time-based data with steady economic factors. The combined hybrid model worked the best, by successfully balancing learning from sequences with keeping things stable. The study did experiments to see what would happen if weather conditions changed, like extreme heat, cold, storms, or dry spells and compared normal forecasts with these changed scenarios to see how demand would shift for products and overall sales. The results show that this new framework not only makes better predictions but also gives useful information on how weather events can affect store sales. By linking prediction with 'what if' analysis, this research moves demand forecasting from just predicting what will happen to helping make better decisions. 2025 IEEE. -
Integrating k-Means++ with ARCANE: A Scalable Framework for Exact Cluster Unlearning
To address the demand for exact data removal in unsupervised clustering, a novel framework for exact machine unlearning is proposed that integrates the K-Means++ algorithm with ARCANE. This framework combines high-quality cluster initialization with targeted partitioning, allowing a more efficient method for removing data without the need for a naive retraining of the model. The proposed model is compared to a SISA-based approach against synthetic and Iris datasets. The ARCANE K-Means++ model demonstrated superior clustering quality, achieving a Silhouette Score of 0.841 to the baseline's performance of 0.263. ARCANE framework also demonstrated better speedup and predictable unlearning times for typical deletion requests than the SISA model. This is a strong, scalable, and provably-exact method for machine unlearning, providing a new and intuitive framework for developing privacy-preserving AI. 2025 IEEE. -
Matrix-Based Apriori Methods for Frequent Pattern Mining: An In-Depth Survey
Data Mining identifies intriguing, useful, and previously unknown patterns and correlations between data stored in databases or warehouses. Frequent Pattern Mining (FPM) is one of the vital methods in the prospering arena of data mining (DM), and it describes the relationship between the items in the datasets. In the last two decades, many studies were carried out in FPM using the Apriori algorithm. The Apriori algorithm requires many database scans and produces numerous candidate itemsets, increasing I/O cost and decreasing computational efficiency. To address these issues, researchers contributed many improved versions of Apriori and proved that those algorithms scan the database only once and identify the frequent itemsets quickly, especially when the itemsets are higher, and provide higher efficiency and feasibility. This research article summarizes matrix-based Apriori algorithms in the literature used for identifying frequent itemsets. 2025 IEEE. -
Flowing Blood Analysis & Separation: Integrating Raman Spectroscopy & Acoustophoresis in a Microfluidic System
The proposed system combines Raman Spectroscopy and Acoustophoresis in a microfluidic environment to provide label-free analysis and separation of the cells present in the blood, including red blood cells, white blood cells and platelets. Based on the Raman Effect, Raman Spectroscopy captures crucial information about the structural and vibrational traits of the blood sample, exposing the molecules to monochromatic radiation and recording their Raman shift. This data is processed using LSTM recurrent neural networks, creating a deep-learning framework for species identification and quantification. The model can forecast the cell population in the blood sample and obtain information on the size and compressibility of individual cells due to the microfluidic nature of the data received by the Raman Spectroscope. Raman spectra allow estimation of the cell size and compressibility, guiding cell separation using acoustophoresis and allowing adjustments of the acoustic wave frequency based on the estimated size and compressibility of the cells. The targeted cells are directed toward a designated collection chamber within the microfluidic system by dynamically adjusting the acoustic wave frequency. Repeated application allows complete separation of the cells and pressure-driven delivery of the targeted cells to their respective collection reservoirs. 2025 IEEE. -
Exploring Emerging AI and IoT Technologies to Enhance Customer Relationship Management in 5G Networks
The convergence of artificial intelligence, Internet of Things for management of 5G networks to revolutionize customer relationship management. It enables customer engagement, efficiency and levels of personalization. The study explores the emergence of IoT and AI to enhance CRM framework within low-latency and high-speed infrastructure provided using 5G technology. Real time data collection with IoT and AI driven data analytics that offer actionable insights to enable customer behaviour, enabling adaptive and predictive CRM strategies. Additionally, enhanced connectivity with 5G technology to facilitate integration of smart devices to ensure interactive and consistent experience for the customers. The present study examined the operational and technical synergy between the technologies and has high impact on the efficiency of CRM, implementation challenges that include data security and privacy concerns. 2025 IEEE. -
AI and IoT in Digital Marketing: Enhancing Automation, Personalization, and Consumer Interaction
An interconnected system of Internet-enabled devices that can collect and transmit statistics via a wireless connection without the assistance of people is known as the Internet of Things (IoTs). Accordingly, the IoT's seductive power is causing significant shifts in the current corporate environment. The digital marketing industry stands to gain the most from this innovation, which is currently causing major shifts in many other sectors. Using a variety of digital marketing strategies, this innovation gathers numerous types of consumer statistics. IoT technological advancements' impact on digital marketing tactics and customer interaction has emerged as a crucial research topic as it begins to pervade many facets of everyday existence. This investigation examines the application of IoT-based machine learning (ML) in digital marketing for the food business. To give ML-based suggestions, consumer data is analyzed, interests are identified, and conduct is predicted using ML approaches. The ensemble technique aggregates the results of multiple ML techniques to produce an individual forecast. The accuracy matrices graphs for the K-nearest neighbor and decision trees produced excellent estimations, with 100% accuracy and 0.0 error, correspondingly. The nae Bayes method achieved 97.2% accuracy with a 0.029 error, successfully identifying the right tags across every category. The guided ensemble of 3 ML methods is demonstrated by effectively enhancing digital marketing tactics in the food distribution industry by reducing duration and expenses. 2025 IEEE. -
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.