This dataset tests whether a model can estimate how close a patient is to a clinical collapse boundary.
The task is not to identify whether collapse has already occurred.
The task is to classify remaining margin before ordinary monitoring or support may no longer be enough.
v0.2 adds counterfactual and adversarial cases.
Some rows have the same MAP, oxygen requirement, respiratory rate, and lactate but different boundary… See the full description on the dataset page:
https://huggingface.co/datasets/ClarusC64/clinical-boundary-distance-v0.2.
