Heart disease prediction using hybrid TabNet architecture with stacked ensemble learning

Area of Science:

• Computational biology
• Medical informatics
• Machine learning in healthcare

Background:

• Cardiovascular diseases (CVDs) are a leading global cause of mortality.
• Current CVD prediction tools face limitations due to data noise and modest accuracy.
• Early detection of CVDs is crucial for effective intervention and patient outcomes.

Purpose of the Study:

• To develop an advanced stacked ensemble framework for enhanced cardiovascular disease risk prediction.
• To integrate deep learning (TabNet) and tree-based (XGBoost) models for improved accuracy and interpretability.
• To provide clinicians with a more reliable tool for early CVD detection.

Main Methods:

• A stacked ensemble framework was developed, integrating TabNet and XGBoost.
• Logistic Regression (LR) or Support Vector Machine (SVM) was employed as a meta-learner.
• The model was evaluated on Kaggle and UCI CVD datasets.

Main Results:

• The proposed ensemble model demonstrated superior performance over baseline models.
• Key performance metrics including accuracy, F1-score, precision, recall, ROC-AUC, PR-AUC, and MCC were significantly improved.
• The framework successfully balanced predictive accuracy with model interpretability.

Conclusions:

• Combining deep learning and tree-based models offers a practical advancement in CVD risk prediction.
• The developed ensemble framework supports clinicians in making more reliable decisions for early CVD detection.
• This approach holds significant potential for improving patient outcomes in cardiovascular health.