A Trustworthy Hybrid AI Framework for Heart Disease Prediction Using Clinical Features and Explainable Deep Learning

Heart disease remains one of the major causes of morbidity and mortality worldwide. Early identification of individuals at risk can support timely clinical assessment and appropriate preventive interventions. Machine learning (ML) and deep learning (DL) techniques have demonstrated potential for predicting heart disease using structured clinical data. However, conventional predictive models may lack interpretability, provide poorly calibrated risk estimates, and exhibit inconsistent performance when applied to different patient populations. This paper proposes a trustworthy hybrid artificial intelligence framework that combines conventional machine learning algorithms, a multilayer perceptron (MLP), probability-level fusion, and explainable artificial intelligence (XAI). The framework processes structured clinical features, including age, sex, chest pain characteristics, resting blood pressure, serum cholesterol, fasting blood sugar, electrocardiographic findings, maximum heart rate, exercise-induced angina, and other relevant clinical indicators. A preprocessing pipeline performs data validation, missing-value imputation, categorical encoding, feature scaling, and class-distribution assessment. Random Forest and Gradient Boosting models capture nonlinear relationships and feature interactions, while the MLP learns complex representations from the clinical variables. Their predicted probabilities are combined through a validation-based fusion strategy. SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) are incorporated to explain global feature importance and individual predictions.