Analysis of the performance of feature optimization techniques for the diagnosis of machine learning-based chronic kidney disease

Abstract

Chronic kidney disease (CKD) slowly decreases one’s kidney ability. A machine learning (ML) based early CKD diagnosis scheme can be an effective solution to reduce this harm. The efficiency of ML techniques depends on the selection and use of the appropriate features. Hence, this research analysis several feature optimization approaches along with a max voting ensemble model to establish a highly accurate CKD diagnosis system by using an appropriate feature set. The ensemble model of this research is structured with five existing classifiers. Three types of feature optimization namely feature importance, feature reduction, and feature selection where for each approach two most proficient techniques are analyzed with the mentioned ensemble model. Based on all analysis the research gets a feature optimization technique called Linear discriminant analysis belonging to the feature selection approach provides the most outstanding result of 99.5% accuracy by using 10-fold cross-validation. The results of this research indicate the efficiency of feature optimization for the diagnosis of ML-based CKD.