Diagnosing Coronary Artery Disease with Perfect Machine Learning Methods and SHAP Explainability

Diagnosis of Coronary Artery Disease (CAD)-one of the highest non-communicable diseases that are affecting millions of people in every corner of the world-requires suitable diagnostic instruments that are effective, precise, and easy to understand.The study proposes an early detection machine learning pipeline for CAD based on the Z-Alizadeh Sani dataset.This pipeline consists of domain-specific preprocessing, SMOTE-based class balancing, hybrid feature selection using RFECV and RFE trimming, and evaluation using several classifiers.XGBoost outperformed all models that were employed with a sensitivity of 0.9637 and ROC AUC of 0.9503.To increase the safety of the categorization, a clinically tuned threshold was used.SHAP analysis revealed the key variables to model openness, such as common chest pain, EFTTE, DM, and BMI.The suggested method stands superior with its diagnostic sensitivity and interpretability when compared with existing norms for clinical applicability.