#calibration

Question 1: Single choice

On held-out data, a classifier assigned a probability of about 0.3 to 200 cases, and 90 of them turned out to be positive. What does this say about the model in this score range?

It is underconfident there; positives occur more often (45%) than the predicted 30%1
It is overconfident there; its scores exceed the observed positive rate2
It is well calibrated, because accuracy rather than frequency is what calibration checks3
Its accuracy in this group must be 45%4

Timed test