Enhancing Heart Disease Prediction Accuracy Through a KNN-SVM-Decision Tree Hybrid Ensemble Framework

Abstract

Cardiovascular disease remains one of the leading causes of mortality worldwide, and early, accurate prediction is essential for timely treatment and prevention. Although numerous hybrid ensemble machine learning models have been proposed in the literature to improve heart disease prediction accuracy, their performance varies considerably depending on which base classifiers are combined and how they are integrated. This study proposes a novel hybrid ensemble model that combines K-Nearest Neighbour (KNN), Support Vector Machine (SVM), and Decision Tree (DT) classifiers through a voting-based strategy for coronary heart disease prediction. The proposed model was developed using the Cleveland Heart Disease dataset, with clinical attributes such as age, blood pressure, cholesterol level, and chest pain type, following data cleaning, normalization, feature selection, and hyper-parameter tuning through grid search and cross-validation. To validate its effectiveness, the proposed ensemble was benchmarked against six existing hybrid ensemble models reported in the literature, including combinations such as KNN+RF+LR, DT+RF, SVM+KNN, RF+DT+SVM+LR, and other multi-classifier ensembles, which achieved accuracies ranging from 81% to 92%. Experimental results show that the proposed KNN-SVM-DT ensemble model achieved the highest accuracy of approximately 98%, together with superior precision, recall, and F1-score, outperforming all six existing ensemble approaches considered in this study. These findings confirm that the proposed hybrid ensemble offers a more robust and reliable framework for heart disease prediction and can serve as an effective clinical decision-support tool for early diagnosis and preventive intervention.