DefinePK

Abstract

Chronic kidney disease (CKD) is a serious global health issue that often progresses without symptoms until irreversible damage has occurred. In this study, we develop and evaluate multiple machine learning models to predict CKD status using routine clinical and laboratory features. Using a publicly available dataset of 400 patients, we implement eight supervised classification algorithms, including Decision Tree, Random Forest, AdaBoost, Gradient Boosting, Stochastic Gradient Boosting, XGBoost, Extra Trees, and K-Nearest Neighbors (KNN). The data were preprocessed through imputation, encoding, and normalization, and models were assessed using accuracy, precision, recall, and F1-score. XGBoost and Extra Trees achieved the highest predictive accuracy (98.3%), followed closely by other ensemble methods. Feature importance analyses consistently identified albumin, hemoglobin, blood urea, and serum creatinine as the most predictive variables. Our findings highlight the utility of ensemble learning techniques for accurate and interpretable CKD prediction, suggesting their potential application in clinical decision support tools.
Keywords: Chronic Kidney Disease, Machine Learning, Classification, XGBoost, Feature Importance, Medical Diagnostics, Ensemble Methods, Clinical Decision Support.