A Hybrid Machine Learning and Deep Learning Framework for Heart Disease Prediction

This paper suggests a machine learning and deep learning hybrid model to predict cardiac diseases in a better way with the use of a Cleveland dataset. The methodology integrates classical ML models, Logistic Regression, SVM, Random Forest, XGBoost, with a shallow MLP network, and combines them via stacked ensemble learning. The experimental evaluation, conducted using 5-fold cross-validation, demonstrates that the stacked ensemble model outperforms individual models, achieving an accuracy of 91%, precision of 92%, recall of 90%, and F 1-score of 91%. Comparative analysis shows improved predictive performance over standalone Random Forest (87% accuracy), XGBoost (89 %), and MLP (85 %). Furthermore, SHAP-based interpretability highlights clinically significant features such as chest pain type, resting ECG, and ST depression as major contributors to prediction outcomes. Calibration and uncertainty analysis confirm that the model provides well-calibrated predictions with low variance, enhancing its reliability in clinical decision-making. The proposed hybrid framework offers a scalable, interpretable, and high-performance solution for early detection of heart disease, supporting its potential application in preventive cardiology and intelligent healthcare systems.