Comparative Evaluation of Machine Learning Classifiers for Explainable Heart Disease Prediction Using the UCI Cleveland Dataset: A Decision Tree-centred Framework Integrating SHAP-based Clinical Interpretability Assessment and Deployable Clinical Decision Support System Development

Abstract

Cardiovascular disease remains one of the leading causes of mortality worldwide, necessitating the development of accurate and interpretable predictive systems that support early diagnosis and clinical decision-making. While numerous machine learning models have demonstrated promising predictive capabilities, many operate as black-box systems that provide limited transparency regarding how predictions are generated. This lack of interpretability presents significant challenges in healthcare environments where trust, accountability, and regulatory compliance are essential. This study presents a comparative evaluation of six supervised machine learning classifiers for heart disease prediction using the Cleveland Heart Disease Dataset. The evaluated models include Decision Tree, Logistic Regression, Support Vector Machine, K-Nearest Neighbours, Random Forest, and Gradient Boosting. A comprehensive machine learning pipeline comprising data preprocessing, feature selection, hyperparameter optimization, repeated stratified cross-validation, and performance evaluation was implemented. Explainable Artificial Intelligence (XAI) techniques based on SHAP were integrated to provide both global and local interpretability of model predictions. Experimental results demonstrate that the Decision Tree classifier achieved the highest overall performance, attaining an accuracy of 98.54%, precision of 98.21%, recall of 98.34%, and F1-score of 98.27%. Furthermore, SHAP-based analysis revealed that chest pain type, number of major vessels, exercise-induced angina, maximum heart rate, and ST depression were the most influential predictors of cardiovascular disease risk. The findings indicate that interpretable machine learning models can achieve predictive performance comparable to or exceeding more complex algorithms while maintaining transparency and clinical usability. The study contributes a reproducible framework for explainable cardiovascular disease prediction and demonstrates the feasibility of integrating interpretable machine learning models into clinical decision support systems. The proposed approach offers a foundation for trustworthy healthcare artificial intelligence applications that balance predictive accuracy with explainability.