Heart disease prediction using hybrid TabNet architecture with stacked ensemble learning

Cardiovascular diseases (CVDs) remain the leading cause of death worldwide, and early detection is critical for timely intervention and improved patient outcomes. However, current prediction tools are often limited by noisy, heterogeneous patient data and modest accuracy. To address this challenge, we propose a stacked ensemble framework that integrates: TabNet, a deep learning model that can identify the most relevant clinical features, and XGBoost, a powerful tree-based method known for its robustness. Their outputs are integrated using a Logistic Regression (LR) or Support Vector Machine (SVM) as meta learner, creating a system that balances accuracy and interpretability. Testing on Kaggle and UCI CVD datasets demonstrate that our ensemble consistently outperforms baseline models across accuracy, F1-score, precision, recall, ROC-AUC, PR-AUC, and matthews correlation coefficient (MCC). These results suggest that combining deep learning with tree-based models offers a practical way to improve risk prediction, supporting clinicians in making more reliable decisions for early CVD detection.