<![CDATA[Early detection is important since chronic kidney disease (CKD) is a major public health problem. This study aimed to compare the discrimination, calibration, and classification performance of logistic regression and four machine learning algorithms for predicting CKD using nationally representative survey data from Indonesia. A Cross-sectional study was conducted using data from the 2023 Indonesian Health Survey (n=15,249). The data were split into training and test sets at an 80:20 ratio to develop five models, namely Logistic Regression, Decision Tree, Random Forest, XGBoost, and Artificial Neural Network. Model performance was assessed using AUC, calibration tests, and classification criteria, with survey weights were incorporated into all performance estimates. The discrimination ability of the machine learning algorithms was higher in the training set (AUC of up to 0.867); however, this advantage decreased in the test set (AUCs of 0.756 – 0.801). In terms of performance consistency and the highest test AUC (0.801), logistic regression was found to be more reliable. Logistic regression also exhibited better calibration, whereas the machine learning models showed evidence of overfitting and miscalibration. Overall, machine learning models did not demonstrate superior predictive performance compared with logistic regression for CKD prediction in this population-based study.]]>
Copyrights © 2026
