When drawing an ROC curve for a binary classifier, we vary, say the probability threshold of one class vs the other and get the curve.

However I'm confused what this means in a multinomial case where the classifier returns probabilities of the various classes (and I'm just picking the class with the maximum probability)

Would like to understand what the ROC curve means for a multinomial case.

  • $\begingroup$ ROC curves aren't defined for multinomial classifiers, so your question makes no sense at all to me. Are you actually asking for ways to convert multinomial classifications so they can be used for ROC analysis? $\endgroup$ – Calimo Apr 24 '17 at 7:11
  • $\begingroup$ @Calimo -- I'm pretty sure Ive seen ROC curves for multinomial classifiers -- so no -- I was asking about ROC curves for multinomial classifiers $\endgroup$ – user1172468 Apr 24 '17 at 14:09
  • $\begingroup$ Can you please find an example? $\endgroup$ – Calimo Apr 24 '17 at 14:12
  • $\begingroup$ @Calimo, let me try -- will post later -- I think the ISLR book may have examples $\endgroup$ – user1172468 Apr 24 '17 at 14:19
  • $\begingroup$ @Calimo, just a clarification -- what I suspect they do is they do a 1-vs-all when do plot em -- you can actually see a CV question on this too: stats.stackexchange.com/questions/2151/… $\endgroup$ – user1172468 Apr 24 '17 at 14:21

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.