Feature Selection using Information Gain on the K-Nearest Neighbor (KNN) and Modified K-Nearest Neighbor (MKNN) Methods for Chronic Kidney Disease Classification

Abstract

Purpose: Kidneys has an important role in the human excretory system. Unhealthy kidneys can affect kidney function. It is important to know the symptoms of chronic kidney disease. One data mining technique that can be applied is the classification technique to determine whether a person has chronic kidney disease or not based on the symptoms (attributes) obtained from medical records. The symptoms of chronic kidney disease obtained amount to 24 symptoms or attributes,Methods/Study design/approach: In this research, the classification of chronic kidney disease is performed using the information gain feature selection method and the KNN and MKNN classification methods. The number of data used is 400 data with 2 classes, namely chronic kidney disease (CKD) and non-chronic kidney disease (non-CKD).Result/Findings: Based on the test results, it was found that the hemo (Hemoglobin) attribute has the highest information gain value, which is 0.6255. The best accuracy for the KNN classification method is 96.61%, and for the MKNN method, it is 98%. Novelty/Originality/Value: The purpose of information gain feature selection is to choose features or attributes that significantly influence chronic kidney disease. Keywords: Chronic Kidney Disease, Information Gain, KNN, MKNN