A robust model for early detection of chronic kidney disease leveraging machine learning and data balancing techniques

Abstract

Chronic kidney disease (CKD) requires reliable early screening, yet clinical datasets are often highly class imbalanced, which can bias machine learning models and reduce detection quality. This study presents a unified evaluation of two imbalance mitigation strategies, synthetic minority over-sampling technique (SMOTE), and cost-sensitive learning, across six classifiers: decision tree (DT), K-nearest neighbors (KNN), logistic regression (LR), random forest (RF), support vector machine (SVM), and extreme gradient boosting (XGBoost). Experiments were conducted on a public CKD dataset with 1,659 records and 54 features using a consistent pipeline including preprocessing, feature selection, imbalance handling, and stratified k-fold cross-validation. Models were assessed with accuracy, precision, recall, and F1-score. Results show that the imbalance strategy materially changes model behavior: cost-sensitive learning generally improves precision, while SMOTE more often increases recall and F1-score. The best overall performance was achieved by XGBoost with cost-sensitive learning, reaching 93% accuracy and 92% precision, outperforming prior reports on the same dataset. RF remained stable across both strategies, whereas KNN was sensitive to SMOTE induced distribution shifts. These findings provide practical guidance for selecting imbalance handling methods to improve healthcare machine learning for CKD detection.