Heart Disease Prediction Using Classification (Naive Bayes)

Abstract

This paper aims toward a greater idea and utilization of machine learning in the medical sector. In this paper, comparative performances of six classification models are presented, when used over the University of California Irvine’s (UCI) Cleveland Heart Disease Records to predict coronary artery disease (CAD). At first, all the 13 provided independent features were used to build the models. On comparing the accuracy of models, it was found that K-nearest neighbors (KNN), support vector machine (SVM), and Naive Bayes have expected and better performances. Thereafter, feature selection is applied to improve prediction accuracy. The backward elimination method and filter method based on the Pearson correlation coefficient is used to choose major predicting features. The accuracy of models using all features and using features selected significantly enhanced the performance of Naive Bayes and random forest, while the other models did not perform as expected. Naive Bayes produced an accuracy of 88.16% on the test set thereafter.