SmartHeart: A conceptual framework for explainable machine learning in cardiovascular risk prediction.

Area of Science:

• Cardiology and Artificial Intelligence
• Computational Health Informatics

Background:

• Cardiovascular diseases (CVDs) are the leading global cause of mortality.
• Early prediction and intervention are crucial for managing heart disease.
• Existing risk prediction models often lack interpretability.

Purpose of the Study:

• To propose the SmartHeart framework for interpretable cardiovascular risk prediction.
• To integrate structured clinical data with a real-time data acquisition pipeline.
• To evaluate the performance of various machine learning models for CVD risk assessment.

Main Methods:

• Utilized a Kaggle heart disease dataset with 11 clinical variables and 1190 records.
• Trained and evaluated six supervised machine learning models: SVC, Random Forest, XGBoost, CatBoost, AdaBoost, and Extra Trees.
• Employed stratified nested 5-fold cross-validation and applied SHAP and LIME for model interpretability.

Main Results:

• Random Forest demonstrated superior performance with 92.86% accuracy and 97.14% AUC.
• SHAP and LIME identified key predictive features: chest pain type, ST slope, and maximum heart rate.
• Statistical tests confirmed the Random Forest model's superiority.

Conclusions:

• The SmartHeart framework provides a foundation for explainable cardiovascular risk prediction.
• The Random Forest model shows significant potential for accurate CVD risk assessment.
• Future integration into cloud-based systems can enable proactive and interpretable healthcare.