Integrating Machine Learning and Deep Learning Approaches for Accurate Cardiovascular Disease Prediction from Electronic Health Records
Natchathirah V. S, T M Meera, S. Padma Devi · 2025
Cardiovascular diseases (CVDs) are a leading cause of death globally, making early prediction and identification of high-risk patients critical for improving treatment outcomes. This study examines the effectiveness of various machine learning and deep learning algorithms in predicting CVD using electronic health records (EHRs). We evaluate the performance of several ensemble methods, including Gradient Boosting, CatBoost, Random Forest, AdaBoost, and LightGBM (LGBM), alongside deep learning architectures such as Long Short-Term Memory (LSTM) and TABNET, utilizing a large EHR dataset. Our results reveal that CatBoost outperforms other ensemble methods, while TABNET achieves the highest accuracy among the deep learning models assessed. Furthermore, we investigate feature importance and the influence of hyperparameter tuning on model performance, providing insights into optimizing predictive capabilities. This research underscores the potential of advanced algorithms in enhancing early detection of CVD and supporting clinical decision-making.