Explainable Hybrid Artificial Intelligence Boosting–Shapley Framework for Cardiac Disease Diagnosis

Abstract

Heart disease is the leading cause of illness and death worldwide. Thus, there is an increasing demand for developing appropriate diagnostic techniques. This paper presents an explainable artificial intelligence hybrid model using a combination of SHapley Additive ExPlanations and Extreme Gradient Boosting to perform interpretable feature explanation and heart disease prediction. In this context, a dataset including 4,240 patient records with 15 clinical parameters was employed to validate the proposed model. XGBoost hyperparameters were optimized for the best setting using a grid-based search combined with 10-fold cross-validation. The conducted experiments showed that the proposed approach outperforms traditional classifiers dealing with the same data. Thus, the F1-score, AUC, recall, accuracy, and precision of the proposed methodology are 99.3%, 98.8%, and 0.97, respectively. SHAP analysis provided both local and global feature attributions, revealing that cardiovascular risk is strongly linked to age, systolic blood pressure, smoking status, and cholesterol level. This hybrid design enhances the reliability of predictions and works well in physician decision-support systems for the early detection and treatment of cardiovascular diseases. It also supports clarity in clinical information and renders it more trustworthy.