Early Detection of Chronic Kidney Disease Using SVM and Logistic Regression Models

Abstract

Chronic Kidney Disease (CKD), a global 
health issue that gradually diminishes kidney 
function, typically goes undiagnosed owing to its lack 
of symptoms. Delays in diagnosis might cause renal 
failure and cardiovascular issues. A dual-model 
machine learning approach for early CKD prediction 
and automated stage categorization utilizing regular 
clinical indicators is presented in this research. The 
suggested system compares model outputs using 
Support Vector Machine (SVM) with Radial Basis 
Function (RBF) kernel and Logistic Regression (LR) 
to increase prediction reliability. 
The system analyzes 24 UCI CKD dataset clinical 
parameters, including serum creatinine, blood 
pressure, hemoglobin, albumin, blood glucose, and 
hypertension. Missing value imputation, Label 
Encoding, and StandardScaler normalization increase 
data consistency and model performance. A Flask
based web application predicts CKD in real time 
using the learned models and classifier confidence 
ratings. 
Beyond binary CKD identification, the method 
calculates the estimated Glomerular Filtration Rate 
(eGFR) using the CKD-EPI algorithm and classifies 
patients into KDIGO stages 1–5. Experimental 
findings show that SVM has 90.5% accuracy, 
whereas Logistic Regression delivers solid 
probability-based clinical predictions. The suggested 
paradigm for early CKD screening and severity 
evaluation in healthcare is low-cost, accessible, and 
clinically helpful.