Heart Disease Prediction using a Stacked Ensemble of Supervised Machine Learning Classifiers

Globally, heart disease related complications cause thousands of deaths each year. Therefore, if diagnostic methods are developed for early and accurate heart disease diagnosis, it can be of great help as far as the survival rate of afflicted patients is concerned. In this research, an ensemble of individual machine learning classifiers (base learners) called StackingCVClassifier is proposed for enhancing the diagnostic performance of these base learners. The developed ensemble learning classifier is made from 3 constituent machine learning models-Logistic Regression classifier, K Nearest Neighbors classifier and Naïve Baye’s classifier. The dataset from Cleveland Clinic Foundation (CVF) is first preprocessed by using various data mining and feature selection techniques to make it more suitable for accurate disease prediction. The proposed StackingCVClassifier ensemble model shows promising results in predicting the occurrence of heart disease in patients. When the ensemble model is employed on the test data, it shows a prediction accuracy of 90.0%, thereby lifting the accuracies of the constituent models which showed accuracies of 86.66%, 88.33%, and 86.66% respectively. Therefore, the proposed stacked ensemble classifier made using Logistic Regression, K Nearest Neighbors, and Naïve Baye’s classifiers as base learners proves to be highly efficient and can, therefore, be used to assist health care practitioners in the overall setup of medical diagnostic systems and processes. These models can be especially useful in those areas wherein the health care is not fully developed to cater to the needs of all people. Additionally, such models provide a noninvasive mode of disease diagnosis as opposed to conventional medical testing which can often times be very invasive.