Improved Detection and Prediction of Chronic Renal Disease by Evaluating Machine Learning Algorithms with Predominantly Reduced Features

Abstract

The kidneys play as an important organ which help in removal of toxic waste from the body. Their malfunctioning may lead to Chronic Renal Disease (CRD) if not attended and treated appropriately at the right time. This chronic situation expedite kidney failure and in-turn death if not diagnosed and attended on time. This work depicts the appropriate, relevant and correlated attributes among all the attributes and reduction of features in the dataset using Chi-square test on to the patients’ dataset for extraction of predominant features for better detection and prediction of CRD. The DPCRD algorithm is implemented and the results are predominantly used in Logistic Regression, K-Nearest Neighbour classification and Random Forest classification technique to enhance and improve its prediction accuracy on CRD. The results revealed that the Random Forest algorithm achieved an improved accuracy of 100.00% after the selection and reduction of attributes using Chi-square test when compared to 99.33% and 99.17% of Logistic Regression and K-Nearest Neighbour classifiers respectively.