Heart Disease Prediction Using a Stacked Ensemble Learning Approach

This study introduces a stacked ensemble machine learning approach to enhance the accuracy of heart disease prediction. The approach begins with a variational autoencoder (VA) for unsupervised learning to identify key patterns in the data. For prediction, several models are applied, including K-nearest neighbors (KNN), support vector machine (SVM), decision trees, Naive Bayes, and quadratic discriminant analysis. Each model offers unique contributions: KNN identifies similar cases, SVM separates data groups, decision trees follow clear decision paths, and Naive Bayes provides probabilistic predictions. Ensemble methods like bagging and AdaBoost further improve accuracy by combining the strengths of multiple models. A dense neural network (DNN), integrated with the VA, captures more complex data patterns. The predictions from all models are stacked and used to train a meta-learner for the final prediction. The proposed heart disease prediction model (HDPM) is the stacked ensemble model that has achieved an accuracy of 92.3% and a precision of 89.3% after 10 training epochs, surpassing existing methods in terms of efficiency and robustness. This approach shows great potential for clinical application, enabling earlier interventions and improving patient outcomes.