Machine Learning Models for Predicting Stroke Risk Among Patients with Coronary Heart Disease

Abstract Stroke is a major global health burden and a frequent severe complication among patients with coronary heart disease. Early identification of individuals at high risk is essential for prevention; however, conventional clinical models often fail to capture the complex interactions underlying stroke risk in this population. This study developed an integrated machine learning framework to predict stroke risk using a large real-world dataset. Multiple algorithms were evaluated, including logistic regression, decision trees, random forests, support vector machines, naive Bayes, multilayer perceptrons, LightGBM, XGBoost, deep learning models, and a stacked ensemble. Class imbalance was addressed using stratified sampling and synthetic minority oversampling. The stacked ensemble demonstrated superior performance, achieving an AUC-ROC of 0.96, precision of 0.91, recall of 0.89, and a Matthews correlation coefficient of 0.82. LightGBM and XGBoost also performed strongly, with AUC-ROC values of 0.94 and 0.95, respectively, and low inference latency. To enhance clinical interpretability, explainable AI techniques (SHAP and LIME) were applied, identifying key risk factors such as prior heart attack, body mass index, age, and lifestyle behaviors. These findings indicate that machine learning models can substantially improve stroke risk prediction in patients with coronary heart disease, supporting clinically actionable and scalable decision-support systems.