Interpretable Machine Learning for Heart Disease Risk Assessment: Leveraging SHAP Values to Identify Clinically Actionable Predictors

Cardiovascular disease remains a leading cause of global mortality, underscoring the urgent need for accurate and interpretable risk prediction tools. This study presents a machine learning framework combining predictive modeling with SHapley Additive exPlanations (SHAP) to identify clinically actionable risk factors for coronary artery disease using the Cleveland Heart Disease dataset. We evaluated three models—Logistic Regression (baseline), Random Forest, and XGBoost—with SHAP-based interpretability to bridge the gap between model performance and clinical utility. Our results demonstrate that XGBoost achieved superior predictive accuracy (88.2% accuracy, AUC(Area under curve=0.91), while SHAP analysis revealed maximum heart rate during exercise (thalach) and ST depression magnitude as the most significant modifiable risk factors, alongside non-modifiable determinants like age and sex. The framework provides physicians with both risk scores and interpretable decision pathways, offering a template for deploying explainable artificial intelligence in preventive cardiology. Key findings highlight the potential of SHAP values to align machine learning outputs with clinical priorities, emphasizing factors amenable to intervention while maintaining diagnostic transparency.