Development and Validation of a Machine Learning Framework for Cardiovascular Disease Risk Stratification Using Clinical, Anthropometric, and Lifestyle Factors

Abstract

Cardiovascular disease (CVD) remains one of the leading causes of mortality worldwide, emphasizing the need for accurate and interpretable predictive models to support early risk assessment and preventive healthcare. This study developed and compared five supervised machine learning algorithms Logistic Regression, Support Vector Machine, K-Nearest Neighbors, Random Forest, and XGBoost for multiclass CVD risk prediction using demographic, anthropometric, biochemical, lifestyle, and clinical variables. A comprehensive preprocessing pipeline, including missing-value imputation, feature standardization, categorical encoding, and elimination of target leakage variables, was implemented before model development. Model performance was evaluated using accuracy, balanced accuracy, precision, recall, F1-score, and receiver operating characteristic area under the curve (ROC-AUC). Among the evaluated algorithms, XGBoost achieved the best predictive performance, obtaining an accuracy of 65.69% and a macro-ROC-AUC of 0.8095, demonstrating superior discrimination compared with the remaining models. Explainable artificial intelligence using SHAP analysis revealed that total cholesterol, HDL cholesterol, age, body weight, smoking status, diabetes status, family history of CVD, physical activity, systolic blood pressure, and fasting blood glucose were the most influential predictors driving model decisions. The alignment between SHAP-derived explanations and established cardiovascular risk factors supports the clinical credibility of the proposed framework. Overall, the findings demonstrate that explainable ensemble machine learning provides a robust and clinically interpretable approach for multiclass cardiovascular risk prediction, offering considerable potential for integration into digital decision-support systems and preventive cardiovascular healthcare.