Predicting heart disease based on influential features with machine learning

Abstract

Heart disease is a major health concern worldwide. The chances of recovery are bright if it is detected at an early stage. The present report discusses a comparative approach to the classification of heart disease data using machine learning (ML) algorithms and linear regression and classification methods, including logistic regression (LR), decision tree (DT), random forest (RF), support vector machine (SVM), SVM with grid search (SVMG), k-nearest neighbor (KNN), and naive Bayes (NB). The ANOVA F-test feature selection (AFS) method was used to select influential features. For experimentation, two standard benchmark datasets of heart diseases, Cleveland and Statlog, were obtained from the UCI Machine Learning Repository. The performance of the machine learning models was examined for accuracy, precision, recall, F-score, and Matthews correlation coefficient (MCC), along with error rates. The results indicated that RF and SVM with grid search algorithms performed better on the Cleveland dataset, while the LR and NB classifiers performed better on the Statlog dataset. Out-comes improved significantly when classification was performed after applying AFS, except for NB, for both datasets.