Global and Local Explainability Analysis of Ensemble Deep Learning Models for Multi-Class Heart Disease Prediction

Abstract

Heart Artificial intelligence techniques are increasingly being used to support medical diagnosis and decision-making processes. Among these techniques, deep learning models have demonstrated strong capability in identifying complex patterns within healthcare datasets. In the domain of cardiovascular disease prediction, such models have shown promising results in identifying high-risk patients. However, one of the major limitations of deep learning systems is their lack of interpretability, since the decision-making process is often difficult to understand. Ensemble learning approaches can improve predictive accuracy by integrating multiple models, but this may further increase system complexity. This study presents an explainable ensemble deep learning framework designed to classify multiple stages of heart disease. The proposed approach integrates SqueezeNet, Improved Long Short-Term Memory (ILSTM), and Bidirectional Gated Recurrent Unit (Bi-GRU) networks. In order to enhance transparency, SHapley Additive Explanations (SHAP) are used to interpret model predictions. Experiments conducted on Cleveland and Switzerland heart disease datasets demonstrate that the ensemble model improves classification accuracy while also providing interpretable explanations of feature influence.