Ensemble learning approach with explainable AI for improved heart disease prediction.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Cardiovascular Medicine

Background:

• Heart disease is a major global health concern, driving the need for accurate and interpretable predictive models.
• Existing models often lack transparency, hindering clinical trust and adoption.
• Explainable AI (XAI) techniques are crucial for understanding model decisions in healthcare.

Purpose of the Study:

• To introduce the Interpretable Ensemble Learning Framework (IELF) for enhanced cardiovascular risk prediction.
• To integrate Explainable Boosting Machines (EBM) with XGBoost, SHAP, and LIME for improved local interpretability.
• To establish a trustworthy benchmark for translational AI in cardiology.

Main Methods:

• IELF was evaluated on the Cleveland (n=303) and Framingham (n=4,240) heart disease datasets.
• Rigorous validation included 5-fold cross-validation, held-out test sets, calibration, and subgroup analyses.
• Explanation stability was assessed using Kendall's τ and Overlap@10 metrics.

Main Results:

• IELF demonstrated robust discrimination, achieving an AUC of 0.899 on Cleveland and 0.696 on Framingham.
• The framework significantly improved recall, F1-score, and AUC compared to EBM on the Framingham dataset (p < 0.05).
• IELF provided transparent feature rankings aligned with known cardiovascular risk factors and stable explanations.

Conclusions:

• IELF is the first framework to combine EBM and XGBoost with SHAP and LIME under strict validation protocols.
• Despite potentially lower headline accuracies than some models, IELF prioritizes reproducibility, interpretability, and clinical reliability.
• IELF serves as a reliable benchmark for AI in cardiovascular risk prediction, balancing performance with transparency.