Machine Learning Framework for Early Detection of Chronic Kidney Disease Stages Using Optimized Estimated Glomerular Filtration Rate

Chronic Kidney Disease (CKD) is a prevalent condition that requires precise diagnosis and staging for effective clinical management. Traditional CKD diagnosis relies on the estimated Glomerular Filtration Rate (eGFR), which is derived from biomarkers such as serum creatinine (SCr) and cystatin C (SCysC). However, the accuracy of eGFR estimation is often limited when applied to diverse patient populations. This proposed study introduces a machine learning (ML)-based framework that integrates regression models for eGFR estimation and advanced classification techniques for accurate CKD staging. The system utilizes Random Forest Regression and Ridge Regression to estimate eGFR, with the latter achieving an impressive accuracy of 0.994. The optimized eGFR values are then used to predict CKD stages via Stacking Classifier and Voting Classifier, with the Stacking Classifier achieving a perfect classification accuracy of 1.0. To ensure model interpretability and support clinical decision-making, Shapley Additive Explanations (SHAP) are used to provide feature importance insights. This ML-based framework offers a scalable, interpretable, and highly accurate solution for the detection and management of CKD, with future work aimed at validating the model with diverse datasets and incorporating additional clinical parameters to further enhance prediction accuracy.

Keywords: Chronic Kidney Disease (CKD), Estimated Glomerular Filtration Rate (eGFR), Machine Learning (ML), Random Forest Regression, Ridge Regression, Stacking Classifier, Voting Classifier, Shapley Additive Explanations (SHAP), Feature Importance, Classification, Detection, Clinical Management.