Coronary artery disease prediction using Bayesian-optimized support vector machine with feature selection.

Introduction: Cardiovascular diseases, particularly Coronary Artery Disease (CAD), remain a leading cause of mortality worldwide. Invasive angiography, while accurate, is costly and risky. This study proposes a non-invasive, interpretable CAD prediction framework using the Z-Alizadeh Sani dataset. Methods: A hybrid decision tree-AdaBoost method is employed to select 30 clinically relevant features. To prevent data leakage, SMOTE oversampling is applied exclusively within each training fold of a 10-fold cross-validation pipeline. The Support Vector Machine (SVM) model is optimized using Bayesian hyperparameter tuning and compared against Sea Lion Optimization Algorithm (SLOA) and grid search. SHapley Additive exPlanations (SHAP) analysis is utilized to interpret the feature contributions. Results: The SVM_Bayesian model achieves 97.67% accuracy, 95.45% precision, 100.00% sensitivity, 97.67% F1-score, and 99.00% AUC, outperforming logistic regression (93.02% accuracy, 92.68% F1-score), random forest (95.45% accuracy, 93.33% F1-score), standard SVM (77.00% accuracy), and SLOA-optimized SVM (93.02% accuracy). Ablation studies and Wilcoxon signed-rank tests confirm the statistical superiority of the proposed model. Discussion: SHAP analysis reveals clinically meaningful feature contributions (e.g., Typical Chest Pain, Age, EFTTE). 95% bootstrap confidence intervals and temporal generalization on an independent test set ensure robustness and prevent overfitting. Future work includes validation on external real-world datasets. This framework provides a transparent, generalizable, and clinically actionable tool for CAD risk stratification, aligned with the principles of network physiology by focusing on interconnected cardiovascular features in predicting systemic disease.