Observed arrival · 2026-10-12
What Medical AI Doesn't Know
An independent research project lets readers inspect how simulated missing clinical features affect model ranking, probability error, and calibration on two public datasets.
- For
- Researchers studying clinical model robustness
- Worth noticing
- At a requested 50% target-feature rate, the page estimates whole-matrix missingness at 8.3% for WDBC and 11.5% for Statlog Heart.
Field notes
The page compares logistic regression, random forest, and histogram-based gradient boosting across WDBC and Statlog Heart, using shared outer splits and controlled MCAR, MAR, and MNAR masks. It distinguishes requested missingness within selected target features from the share of the complete feature matrix removed, and reports ranking, probability error, and calibration alongside saved results. The project says its experiments use public benchmark feature matrices, not patient records submitted through the site.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue