A Study on Performance Improvement of Heart Disease Prediction by Attribute Selection Methods

Abstract

Heart pumps blood for all tissues of the body. The deteriorate of this organ causes a severe illness, disability and death since cardiovascular diseases involve the diseases that related to heart and circulation system. Determination of the significance of factors affecting this disease is of great importance for early prevention and treatment of this disease. In this study, firstly, the best attributes set for Single Proton Emission Computed Tomography (SPECT) and Statlog Heart Disease (STATLOG) datasets were detected by using feature selection methods named RFECV (Recursive Feature Elimination with cross-validation) and SS (Stability Selection). Secondly, GBM (Gradient Boosted Machines), NB (Naive Bayes) and RF (Random Forest) algorithms were implemented with original datasets and with datasets having selected attributes by RFECV and SS methods and their performances were compared for each dataset. The experimental results showed that maximum performance increases were obtained on SPECT dataset by 14.81% when GBM algorithm was applied using attributes provided by RFECV method and on STATLOG dataset by 6.18% when GBM algorithm was applied using attributes provided by RFECV method. On the other hand, best accuracies were obtained by NB algorithm when applied using attributes of SPECT dataset provided by RFECV method and using attributes of STATLOG dataset provided by SS method. The results showed that medical decision support systems which can make more accurate predictions could be developed using enhanced machine learning methods by RFECV and SS methods and this can be helpful in selecting the treatment method for the experts in the field.