Machine Learning Approaches to Identify Significant Features for the Diagnosis and Prognosis of Chronic Kidney Disease

Chronic kidney disease (CKD)is a long-term disease in which the kidneys are failed and needed dialysis or kidney transplantation in the severe stage. Thirty-seven million people in the United States have CKD, and millions more are at risk. Early diagnosis can help prevent kidney disease from progressing to kidney failure. The aim of this study is to make CKD diagnosis predictions based on symptoms or characteristics seen in a particular case, regardless of whether the stage is acute or chronic, and to classify the most important features that contribute to CKD. We illustrate machine learning methods to predict chronic kidney disease using clinical evidence and statistical tests to identify the most important reasons for CKD. Seven machine learning methods are explored including K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Logistic Regression (LR), Decision Tree, Random Forest, Naive Bayes, and XGBboost classifiers to predict CKD and two statistical tests including Student’s T-test and chi-squared test to identify most significant features. The results showed that the tree-based classifier performs best in the kidney diagnostic procedure with accuracy of 100%. In addition,hemoglobin, packed cell volume, specific gravity, red bloodcell countsand albumin are the most predictive biomarker for CKD.