Is it possible to calibrate the probabilities of a binary classifier when the class priors are unknown?

In cases where the data is obtained with selection bias (i.e. more positives than negatives in data collection, but in actuality, there are more negatives than positives as an example), without full knowledge of the proportions, how do we calibrate the model?

This question is also applicable in the case of Positive Unlabeled learning, where if we had limited positives samples, but a variable amount of unlabeled data, how do we calibrate the probabilities of such a model?


Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Browse other questions tagged or ask your own question.