When working with ROC-AUC as a metric for binary classification, one often considers a value of 0.5 as a baseline from a random classifier (i.e. a data-blind classifier that randomly classifies test instances with equal probability).

I have read that average precision (or more generally, mean average precision) may be a better metric when the positive class is of higher interest than the negative class. This claim deserves its own question, but aside from that, what is a reasonable random baseline for (baseline) average precision?

I am inclined to think that such random baseline should be P/(P+N) (i.e. fraction of positives of the total). How should I go about stablishing this baseline?



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.