Predicting Heart-Attack Risk Using Machine- And Deep-Learning Methods In Ethiopia
| dc.contributor.advisor | Melkamu Hunegnaw (PhD) | |
| dc.contributor.advisor | Dejuma Yadeta (PhD) | |
| dc.contributor.author | Emirt Worku | |
| dc.date.accessioned | 2025-12-31T09:46:52Z | |
| dc.date.available | 2025-12-31T09:46:52Z | |
| dc.date.issued | 2025-09 | |
| dc.description.abstract | Heart attack, one of the most severe forms of cardiovascular disease (CVD), remains a leading cause of mortality worldwide. In 2022, CVD accounted for an estimated 19.8 million deaths, representing 32% of all global deaths, with a growing burden in low- and middle-income countries such as Ethiopia. Major risk factors contributing to heart attack include high blood pressure, high cholesterol, diabetes, smoking, obesity, poor diet, and physical inactivity. The early detection and prediction of heart attack risk are critical in reducing mortality. The purpose of the thesis is to predict risk of heart attack in the human body and to provide suggestions to individuals to reduce the risk in the future. In this thesis, we develop a model using machine learning (ML) and deep learning (DL) using a public data set obtained from the Behavior Risk Factor Surveillance System (BRFSS), the world’s largest continuously conducted health survey system, with the help of the Centers for Disease Control and Prevention (CDC) and secondary data from Tikur Anbessa Specialized Hospital (TASH) that include clinical data and demographic data from the person. We evaluated a wide range of classical ML models, including logistic regression (LR), K-Nearest Neighbours (KNN), Decision trees (DT), Random Forest (RF), and Gradient Boosting (GB) to assess performance compared to DL models. The performance of each ML algorithm was evaluated using cross-validation techniques and standardized metrics to ensure reliability. We also evaluate several DL models, including the Feedforward Neural Network (FNN), Wide & Deep model, Residual Network, and Attention-based model. The diversity of DL models explored in our study allows one to capture complex, nonlinear relationships within health data. The public availability of large-scale health data has allowed us to develop computational techniques to improve medical diagnostics and screen the high-risk patient. In this context, our work contributes not only to the development of predictive modeling but also to the development of the graphical user interface (GUI) application, which is designed to be accessible for clinical use and aims to support healthcare professionals. Finally,We found that most DL models achieved fairly similar performance, with the best results showing balanced accuracy (95%), precision (0.95), recall (0.96), F1 score (0.95), and AUC (0.97) in the FNN. This highlights the promise of DL approaches in advancing early diagnosis and personalized care for CVD. Based on these results, the FNN model was integrated into the developm GUI application for the prediction of CVD risk and decision support in real time. | |
| dc.identifier.uri | https://etd.aau.edu.et/handle/123456789/7595 | |
| dc.language.iso | en_US | |
| dc.publisher | Addis Ababa University | |
| dc.subject | Cardiovascular disease | |
| dc.subject | Machine learning | |
| dc.subject | Deep learning | |
| dc.subject | Models | |
| dc.subject | prediction | |
| dc.subject | Ethiopia | |
| dc.title | Predicting Heart-Attack Risk Using Machine- And Deep-Learning Methods In Ethiopia | |
| dc.type | Thesis |