A COMPARISON OF MACHINE LEARNING CLASSIFIERS AND UNCERTAINTY QUANTIFICATION TECHNIQUES FOR PREDICTING HEART DISEASE

Abstract

In all over the world, Heart disease is the leading disease which causes sudden death, so for early detection of this disease and treatment need a reliable and accurate prediction models. This study highlights comprehensive approaches of machine learning classifiers such as Logistics Regression, Naïve Bayes, Random Forest Tree, K-Nearest Neighbor, Decision Tree and Support Vector Machine are applied on a dataset of indication and different symptoms of this disease. The uncertainty quantification techniques to increase the accuracy such as dropout uncertainty, ensemble and quantile regression are used. Also F1 score, accuracy, precision, recall,, support, confusion matrix are computed and the Receiver Operating Characteristic and area under the curve  to evaluate the overall performance of the classifiers are discussed. The highest accuracy of testing is 99.02% with random forest classifier, when integrated with quantile regression it gives same output. Moreover, this study highlights the value of uncertainty quantification to improve the accuracy of many classifiers to predict the heart disease. This study highlights the value of uncertainty quantification in enhancing version reliability and shows how well the Random Forest classifier predicts coronary heart disease. The results provide comprehensive understandings of how to apply machine learning models in academic contexts to improve patient experiences and diagnostic accuracy.