An Intelligent Clinical Decision Support System for Chronic Kidney Disease Detection Using Machine Learning

Abstract

Chronic Kidney Disease (CKD) is a disease that is progressive, and it may cause serious outcomes and renal failure unless identified in its early stages. This paper presents a smart and explainable model on the prediction of chronic kidney disease with the help of machine learning and ensemble algorithms. The study uses CKD data in the UCI machine learning repository which consisted of 400 patient records collected at the Apollo hospital in India with 25 numerical and categorical variables. Statistical and model-based imputation are used to handle missing values, and Synthetic Minority Over-sampling Technique (SMOTE) is employed to address class imbalance (CKD and non-CKD instances). Complete preprocessing, feature engineering, and feature selection methods are used: Mutual Information, Variance Thresholding, Recursive Feature Elimination, and Sequential Feature Selection, which identify six clinically significant features, namely, hemoglobin, serum creatinine, albumin, hypertension, age, and diabetes mellitus. Stratified K-Fold cross-validation is used to test various classifiers, including the Logistic Regression, Gaussian Naive Bayes, Support Vector Machine, Decision Tree, Random Forest and AdaBoost. A hybrid Voting Classifier which combines random forest with adaboost achieves a very high performance with the accuracy, precision and recall of 99.4 and F1-score of 99.3. The interpretation of the model is enhanced by using LIME and SHAP to explain single predictions and contributions of features. A web application based on Flask includes user registration, authentication, secure data entry, real-time predictions, and explained results, delivering the output of the results as either chronic kidney disease detected or no chronic kidney disease detected, which makes it reliable and user-friendly clinical decision support.