Explainable Heart Disease Diagnosis with Supervised Learning Methods

Abstract

The objective of this study is to develop a heart disease diagnosis model with a supervised machine learning algorithm. To that end, random forest (RF), support vector machine (SVM), Naïve Bayes (NB), and extreme boosting (XGBoost) are employed in a medical heart disease dataset to develop a model for heart disease prediction. The performance of the algorithms is investigated and compared for automation of heart disease diagnosis. The best model is selected, and a grid search is applied to improve model performance. The simulation result shows that the XGBoost model outperforms the others, achieving 99.10% accuracy, and receiver operating characteristic curve (AUC score=0.99) compared to RF, SVM, and NB on heart disease detection. Finally, the obtained result is interpreted with Shapley additive model explanation (SHAP) to investigate the effect of each feature on the diagnosis of heart disease. A case study on heart disease diagnosis shows an important insight into the impact of the feature on the diagnosis performance of the supervised learning method. The developed model had an expressively higher prediction accuracy, indicating the utility of supervised learning systems in detecting heart disease in the early stages.