Hybrid Machine Learning Approach for Enhanced Heart Disease Prediction: A Multi-Classifier Consensus Model

Heart disease remains among the major causes of deaths in the world and it accounts several deaths. Finding the cause of heart disease requires a lot of time and effort, especially for physicians and other medical professionals, making the traditional methods for predicting this condition one of the most difficult tasks in the medical industry. This study aimed to address this challenge by introducing a hybrid machine learning designed to transcend these limitations for enhanced heart disease prediction which a multi- classifier consensus model, which leverages the strengths of six machine learning algorithms Logistic Regression (LR), Random Forest (RF), Decision Tree (DT), Support Vector Machine (SVM), AdaBoost, and K- Nearest Neighbors (KNN) to achieve superior predictive accuracy, robustness, and interpretability. The proposed model integrates these algorithms into a Voting Classifier using soft voting, assuring an agreement- predicated prognosticating that benefits from the unique capabilities of each algorithm. The Synthetic Minority Oversampling Technique (SMOTE) is used to rectify class imbalance, optimal feature selection was accomplished using Recursive Feature Elimination (RFE). On the UCI Cleveland dataset, the proposed approach achieved test accuracy of 98.04 %. To enhance interpretability, SHAP (SHapley Additive exPlanations) is employed, revealing strong alignment with known medical risk factors such as ca, thalach and oldpeak. While external validation was attempted on Framingham and Statlog datasets, feature mismatches and class imbalanced limited definitive conclusions. This study offers a high-performance, interpretable, and repeatable approach for clinical decision assistance in cardiovascular risk prediction.