Interpretable Blending Ensemble Learning Models for Cardiovascular Disease Prediction

Cardiovascular disease (CVD) is considered a prominent life-threatening cause of death but early-stage pre-diction can reduce the percentage of heart failure. This study compares parametric and non-parametric Machine Learning (ML) classifiers to predict early-stage CVD and designed a blending ensemble model by combining Multilayer Perceptron (MLP), Random Forest (RF), Support Vector Machine (SVM) namely Blending MRS (B_MRS). Also use explainable AI tools to find the blackbox ML model interpretability, which finds how the features affect the performance of the classifiers. Additionally, this paper employ Logistic Regression, Naive Bayes, and K-Nearest Neighbors in this analysis. In most cases, the parametric algorithms show a better result than non-parametric algorithms, but B_MRS shows superiority by achiving 97.89% accuracy. This research find the feature's importance through SHAP and Shapash that indicates ‘ca’, ‘thaI’, and ‘cp’ are top 3 features. The local explainability clearly indicates that the positive influence of the top features cause CVD, where the positive influence of the less important features have negative impact on CVD. This model could have potential applications in clinical decision-making, supporting early diagnostics and personalized intervention strategies for at-risk populations.