Heart Disease Prediction Using Ensemble Tree Algorithms: A Supervised Learning Perspective

Heart disease stands as a leading cause of morbidity and mortality globally, presenting a significant public health challenge. Therefore, early prediction and detection are critical, leading to timely and appropriate interventions at early stages. Four ensemble tree-based algorithms were used in this study: adaptive boosting, extreme gradient boosting, random forest, and extremely randomized trees, investigating their ability to predict heart disease. Data related to heart disease clinical features was obtained from the open Kaggle Machine Learning Dataset repository. Adaptive Boosting stands out as the highest performer, achieving an average testing accuracy of 93.70%, precision of 93.71%, recall of 93.70%, and F1 score of 93.69%, along with the highest AUC score of 0.9708, across all competing models considered in the study. These metrics indicate a superior ability to distinguish between patients with and without heart disease, effectively making it particularly valuable for clinical applications where early detection can save lives. The SHapley Additive exPlanations (SHAP) framework adopted to investigate the relative importance of the features in predicting heart disease revealed the most influential predictors (ST slope, chest pain type, old peak, and cholesterol), further aiding the understanding of heart disease mechanisms. Future work should explore the integration of ensemble learning algorithms with real-time patient monitoring systems. This integration could allow for continuous health status updates, equipping predictive models with the information necessary to facilitate dynamic, real-time interventions that are more closely aligned with patient needs.