Ensemble learning with explainable AI for improved heart disease prediction based on multiple datasets.

Area of Science:

• Cardiology
• Machine Learning
• Artificial Intelligence

Background:

• Heart disease is a leading global cause of mortality.
• Early detection and accurate diagnosis are critical for effective patient management.
• Machine learning offers potential for improving cardiac condition diagnosis.

Purpose of the Study:

• To enhance heart disease prediction accuracy using ensemble machine learning techniques.
• To compare the performance of stacking and voting ensemble methods against individual models.
• To provide transparency in model predictions using explainable AI (XAI).

Main Methods:

• Trained fifteen base machine learning models on two heart disease datasets.
• Developed ensemble models using stacking (with a meta-model) and voting (majority vote) with six selected base models.
• Conducted statistical validation using Friedman aligned ranks test and Holm post-hoc analysis; incorporated SHAP for XAI.

Main Results:

• Ensemble models, especially stacking, demonstrated superior performance over individual base models.
• Achieved higher accuracy and improved predictive outcomes in heart disease classification.
• SHAP analysis provided insights into feature influence on prediction, enhancing model interpretability.

Conclusions:

• Stacking and voting ensemble methods significantly enhance heart disease prediction performance.
• These ensemble approaches represent a valuable tool for clinical decision-making in cardiology.
• Explainable AI integration increases trust and understanding of machine learning-based diagnostic tools.