XSG-Ensemble: A Weighted Voting Ensemble Machine Learning Model for Enhanced Chronic Kidney Disease Prediction

Abstract

Chronic Kidney Disease (CKD) affects over 800 million people worldwide, causing 5-11 million annual deaths and imposing significant socioeconomic burdens through comorbidities such as cardiovascular disease and end-stage renal disease (ESRD). Early detection is crucial to mitigate disease progression, yet conventional diagnostics often detect damage only after onset. This study proposes XSG-Ensemble, a novel weighted ensemble machine learning model integrating Support Vector Machine (SVM), XGBoost, and Gradient Boosting via optimized soft voting, to enhance CKD prediction accuracy and robustness. Using a dataset of 2,286 patients with demographic, lifestyle, clinical, and laboratory features (e.g., GFR, serum creatinine, proteinuria), rigorous preprocessing was applied, including median/mode imputation, IQR outlier capping, label encoding, and SMOTE for class balancing. The dataset was split 80/20 for training/testing, base models were hyperparameter-tuned via GridSearchCV, and ensemble weights were optimized similarly. XSG-Ensemble achieved 98.25% precision, recall, F1-score, and test accuracy, with 97.76% cross-validation accuracy and 99.85% ROC AUC, outperforming base models and prior benchmarks by 3–8%, and demonstrating only 3 misclassifications in 342 test instances. This framework highlights the potential of ML-driven nephrology for early CKD intervention, reducing ESRD risk, healthcare costs, and mortality, with future validation on multi-center cohorts expected to further confirm its clinical utility.