Comparative Evaluation of Machine Learning and Deep LearningApproaches for Heart Disease Prediction

Abstract

Cardiovascular disease (CVD) remains the leading cause of death globally, responsible for an estimated 20.5 million fatalities in 2025 according to updated World Health Organization projections. The critical importance of early, accurate, and scalable diagnostic systems has intensified interest in machine learning and deep learning methodologies for automated heart disease prediction. This paper presents a comprehensive 2026 benchmark study evaluating six machine learning and deep learning algorithms applied to the Cleveland Heart Disease data-set from the UCI Machine Learning Repository. The algorithms evaluated are Logistic Regression, Decision Tree, Random Forest, K-Nearest Neighbor, Gradient Boosting (XGBoost), and a Feed-Forward Neural Network. All models are assessed using 10-fold stratified cross-validation with hyper-parameter optimization, producing robust and reproducible performance estimates. Updated results demonstrate that the Gradient Boosting model achieves the highest test accuracy of 91.80%, followed closely by Logistic Regression at 88.52% and Random Forest at 85.25%. Comprehensive evaluation metrics including precision, recall, F1-score, AUC-ROC, and Matthews Correlation Coefficient (MCC) are reported for clinical interpretability. Additionally, SHAP (SHapley Additive exPlanations) feature importance analysis is introduced to identify the most diagnostically significant attributes. Benchmarking against literature published from 2021 through 2025 contextualizes these results within the current state of the art.