Explainable ensemble model reports stronger heart-disease prediction on public datasets

The proposed MCDELE system adds cluster-derived features and statistical meta-features to a two-tier stacking ensemble. On an aggregated public Kaggle/UCI dataset, the authors report 90.22% accuracy, 91.26% F1 and 93.77% AUC, with SHAP explanations and Wilcoxon testing.

Why it matters: The work combines predictive performance with interpretability, but public retrospective data does not establish clinical utility, fairness or prospective generalisation.

Paper: Scientific Reports