Augmenting heart disease prediction with explainable AI: A study of classification models

Abstract

Although heart disease stands as a prominent contributor to worldwide deaths, not all individuals affected by it ultimately fall prey to its effects. Timely diagnosis and effective treatment can offer those with heart conditions a high-quality life in their later years. Consequently, early disease detection using accessible medical data has been a central goal for researchers in recent decades. Traditionally, researchers relied on statistical tools for this purpose. However, machine learning algorithms, especially classification models, have gained prominence with the growing accumulated data. These algorithms have shown promise in predicting heart disease based on individual data. Our study employed various classification algorithms to predict heart disease incidence using the available dataset. We prioritized model reliability by incorporating the conformal classifier. Our results have shown that boosting algorithms, such as XGBoost and CatBoost, demonstrated exceptional performance with promising metrics. These models identified chest pain type and ST segment slope as crucial indicators of heart disease. Boosting algorithms exhibited a compelling combination of broad coverage and a small prediction set size, making them well-suited for heart disease prediction. Furthermore, we employed explainable artificial intelligence-boosting algorithms to enhance the interpretability of our predictions.