Evaluating Reliability in Machine Learning Models for Early Chronic Kidney Disease Prediction: A Systematic Review of Data Leakage and Predictor Stability

This systematic review examined 19 studies using interpretable machine learning for chronic kidney disease prediction. It proposes a taxonomy and quantitative scoring framework for information leakage. Studies classified as high leakage reported an average accuracy of 95.48%, compared with 80.2% for leakage-free studies, a difference of about 15.28 percentage points. Cross-study feature analysis also found that more than 80% of predictors lacked reliable reproducibility. The authors argue that some reported gains may reflect methodological flaws rather than genuine predictive capability.