HEART DISEASE PREDICTION AND PREVENTION USING DEEP LEARNING

Abstract

Heart disease remains a leading cause of morbidity and mortality worldwide, underscoring the urgent need for accurate and early risk prediction. This thesis explores the use of advanced deep learning methods to improve the identification and prevention of heart disease. Utilizing publicly available clinical datasets, the study systematically addresses challenges such as class imbalance, feature selection, and model transparency. Several neural network architectures— including Convolutional Neural Networks (CNN), Bidirectional Long Short-Term Memory (BiLSTM), and hybrid CNN-LSTM models—are implemented and assessed through stratified cross-validation. To counteract the effects of imbalanced data, the Synthetic Minority Oversampling Technique (SMOTE) is integrated into the workflow, resulting in measurable gains in model performance. Among the tested architectures, the CNN-BiLSTM consistently delivers the highest accuracy, F1-score, and ROC-AUC, demonstrating the value of combining spatial and temporal feature extraction. To ensure clinical relevance, interpretability tools such as SHAP and LIME are applied, revealing key risk factors and supporting individualized prevention recommendations. The findings suggest that the proposed deep learning framework not only advances predictive accuracy but also provides actionable insights, paving the way for its adoption in real-world healthcare settings to support proactive cardiovascular care.