Ensemble Machine Learning Frameworks for Coronary Artery Disease Classification: A Systematic Review

Coronary Artery Disease (CAD) is the leading cause of global cardiovascular mortality, requiring reliable, non-invasive clinical decision support systems as alternatives to costly coronary angiography. Although standalone machine learning (ML) classifiers are the core of automated risk assessment, they exhibit high variance for small cohort data, sensitivity to class imbalance, redundant features and unstable decision boundaries. This review is a systematic assessment of the ability of ensemble learning and algorithmic preprocessing to address these limitations. After screening a total of 1,284 articles from Scopus indexed papers in the period of 2021-2026, we analyze and compare 20 representative papers from four categories: (1) feature selection, evolutionary optimization and class-balancing pipelines; (2) homogeneous bagging and gradient-boosting architectures; (3) heterogeneous voting ensembles; and (4) multi-level stacking and hybrid frameworks. Critically, our analysis reveals a methodological paradox: accuracies reaching 99% to 100% on small benchmark datasets frequently stem from pre-split Synthetic Minority Oversampling Technique (SMOTE) data leakage and benchmark saturation rather than true clinical generalizability. Ultimately, clinical translation requires strict out-of-sample validation, computational efficiency for low-resource settings, and quantitatively audited Explainable Artificial Intelligence (XAI).