A decision support system approach for chronic kidney disease diagnosis using machine learning

Abstract

Purpose: Chronic kidney disease (CKD) is a progressive condition in which early detection is essential for improving patient outcomes and reducing healthcare burden. This study purpose a machine learning based approach as a clinical decision support mechanism to assist in the early identification of CKD using routinely collected patient data. Method: A benchmark dataset comprising 400 patient records with demographic, laboratory, and clinical features was utilized. A structured preprocessing pipeline was implemented to address missing values and heterogeneous data types through imputation, normalization, and encoding. Three supervised learning models, Logistic Regression, Random Forest, and K-Nearest Neighbors (KNN) were evaluated using a stratified 80:20 train test split, with performance assessed via accuracy and F1 score. Finding: All models demonstrated strong predictive performance, with Logistic Regression achieving perfect classification (accuracy = 1.000, F1-score = 1.000). Feature analysis highlighted clinically relevant biomarkers, including hemoglobin, packed cell volume, specific gravity, and albumin, as key predictors of CKD. Conclusion: The results suggest that machine learning models, particularly when integrated into a decision support system, can support clinicians in early CKD detection and risk assessment. Such systems have the potential to enhance diagnostic accuracy, facilitate timely intervention, and improve health management processes.