Artificial Intelligence-Assisted Early Detection of Chronic Kidney Disease: A Comparative Machine Learning Approach
TL;DR
The results reinforce the potential of ensemble-based machine learning to support early, data-driven CKD screening and highlight the biomarkers most relevant to future regional screening tools.
Abstract
Chronic Kidney Disease (CKD) is a progressive condition that frequently develops without early symptoms, making timely detection critical for preventing irreversible renal damage, particularly in patients with diabetes and hypertension. This study evaluates the performance of three supervised machine learning models Logistic Regression, Random Forest, and Support Vector Machine (SVM) for early CKD classification using a clinical dataset of 400 patient records comprising demographic, clinical, and biochemical attributes such as serum creatinine, hemoglobin, packed cell volume, and blood pressure. Following data cleaning, missing-value imputation, and feature scaling, models were trained on an 80/20 stratified split and evaluated using accuracy, sensitivity, specificity, precision, F1-score, and ROC-AUC. The Random Forest classifier achieved the strongest performance (100% test accuracy, ${A} {U} {C}={1.00})$, with Logistic Regression and SVM close behind (98.75% accuracy each). Feature importance analysis identified hemoglobin, packed cell volume, serum creatinine, and specific gravity as the dominant predictors of CKD status. While the near-perfect scores reflect the strong separability of this widely used benchmark dataset rather than clinical-grade generalizability, the results reinforce the potential of ensemble-based machine learning to support early, data-driven CKD screening and highlight the biomarkers most relevant to future regional screening tools.