Boosted classifier and features selection for enhancing chronic kidney disease diagnose

Abstract

Chronic kidney disease (CKD) is a disease caused by degeneration function of the kidneys. CKD is top ten leading causes of death in the world. There are two leading causes of CKD, i.e. diabetes and hypertension. When the symptoms become more severe, the disease can only be treated with dialysis and kidney transplantation. This disease can be treated if the diagnose is conducted appropriately and quickly. However, the signs and symptoms are often not specific. Because of that, diagnosis from medical personnel is may subjective and vary. This study developed machine learning method using ensemble learning and feature selection to improve the quality of CKD diagnosis. The CKD dataset was taken from UCI machine learning repository, it contain 400 instances. CKD dataset have 24 attributes including signs, symptoms and risk factors that might appear due to CKD. In this study, features were selected using a Correlation-based Feature Selection (CFS) and AdaBoost was used for ensemble learning to improve the detection of CKD. K-Nearest Neighbour algorithm (kNN), Naive Bayes and Support Vector Machine (SVM) was used as base classifier. Overall, the best result was achieved by combination of kNN classifier with CFS and AdaBoost, with 0.981 accuracy rate, 0.980 recall rate and 0.980 f-measure rate. Highest precision rate was achieved by the combination of Naive Bayes classifier with CFS and AdaBoost, with 0.981 precision rate.