Machine learning CKD prediction models suffer from data leakage and unstable predictors

A systematic review of machine learning models for early chronic kidney disease (CKD) prediction has revealed significant issues with data leakage and predictor stability. The review analyzed nineteen studies, introducing a taxonomy and scoring framework to evaluate information leakage. Studies with high leakage reported an average accuracy of 95.48%, substantially higher than the 80.2% accuracy reported by leakage-free studies, indicating inflated performance metrics. Furthermore, the analysis found that over 80% of predictors lacked reliability, suggesting that reported performance gains are often due to methodological limitations rather than true predictive capability. AI

IMPACT Highlights critical methodological flaws in ML for healthcare, suggesting current performance metrics may be unreliable and impacting trust in AI-driven diagnostics.

RANK_REASON This is a systematic review paper published on arXiv, detailing methodological issues in machine learning for healthcare. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• arXiv
• Chronic Kidney Disease
• interpretable machine learning
• machine learning
• Predictor Stability

• paper
• safety
• product

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →