An Explainable Hybrid Machine Learning Framework for Proactive Heart Disease Prediction and Risk Classification

Abstract

Traditional clinical cardiovascular risk scoring systems are based mainly on linear risk factors that often do not adequately reflect the non-linear interactions of a heterogeneous demographic and physiological risk factors. In this study, an explainable hybrid machine learning system for early heart disease prediction and multistage risk categorization is presented. The framework integrates these two types of representation learning and utilizes tree-based ensemble methods to produce not only a binary diagnosis but also a multi-class risk classification. A post-hoc Explainable Artificial Intelligence (XAI) is incorporated into the pipeline to provide local and global feature attributions that help to explain the limits of the decisions made per prediction. Evaluation with standardized clinical benchmarks shows that the hybrid method exceeds the performance of the baseline models using only a single model on certain parameters (AUC-ROC, F1, sensitivity and specificity). The framework provides a practical decision support tool for early cardiovascular screening and targeted clinical intervention, with high predictive fidelity and clinician-friendly, understandable rationales.