I built a multiclass (and very imbalanced classes) classifier. When evaluating I found an average F1 of .98 and the classifier seems to be working rather well. However, on evaluating the ROC and Precision/Recall (PRC) curves, I find that it's a rather bad performing model.

enter image description here

I plotted the ROC and PRC for one class and the AUC are .52 and .1 respectively. However, the F1 for this class is .99. So I understand that the evaluating with F1 only considers if the model picked the correct class. But am I right in interpreting that according to the curves, the probability output of the classes for each instance is near uniform? Even if the highest probability was picked, the probability of the negative classes are close to that of the correct class? Is that how I should interpret these curves?


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.