Cardiovascular screening model accuracy attributed to data leakage, not model type

A new study published on arXiv investigates the accuracy of cardiovascular screening models, revealing that reported high accuracy is primarily due to target leakage rather than the model's class. Researchers found that removing just two post-diagnostic features significantly reduced the AUROC for all tested models, collapsing them into a narrow performance band. The study highlights that evaluation practices, not model capacity, are the main constraint, and emphasizes the benefits of transparency for auditing fairness and uncertainty. AI

IMPACT Highlights the critical need for robust data validation in AI model development, particularly in sensitive domains like healthcare.

RANK_REASON The cluster contains a research paper published on arXiv detailing a new audit of machine learning models. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• alphaXiv
• Behavioral Risk Factor Surveillance System
• CatalyzeX
• DagsHub
• Explainable Boosting Machine
• Gotit.pub
• Hugging Face
• Piet Mondrian
• Raad Bin Tareaf
• ScienceCast

• paper
• other

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