Hybrid Learning Framework for Explainable Cardiovascular Disease Detection

Cardiovascular disease (CVD) is a leading cause of global mortality, necessitating predictive models that are both accurate and interpretable. This study introduces the Hybrid Polynomial Ensemble Model (HPEM), a novel ensemble learning framework that combines Random Forest (RF), Gradient Boosting (GB), and Support Vector Machines (SVM), with Extreme Gradient Boosting (XGBoost) as a meta-learner. The model integrates advanced feature engineering, including second-degree polynomial expansion and mutual information-based selection, to enhance learning from heterogeneous patient data. HPEM was evaluated on three publicly available datasets, achieving an accuracy of 91%, 88%, and 89% respectively. It also attained F1-scores of 0.91, 0.88, and 0.89, and Area Under the Curve (AUC) values up to 0.96, outperforming baseline models across all metrics.To ensure transparency, explainability tools such as SHapley Additive exPlanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME) were applied, identifying key clinical predictors including cholesterol levels, exercise-induced angina, and age. The model also achieved Cohen’s Kappa scores above 0.79 and Matthews Correlation Coefficient (MCC) values above 0.80, demonstrating reliability and robustness. These results underscore the potential of HPEM as a clinically relevant tool for early detection and risk stratification of cardiovascular disease.