Search code examples
classificationrocauc

Overfitting and ROC curves

Is it possible to understand if a classifier is overfitted just by looking at its ROC curve? I see that if its AUC it's too high (example 98%) is likely to be overfitted, but it can also mean that the classifier is just really good. Is there a way to tell these two cases apart?

Solution

  • Short answer: no, you can't.

    Long answer: to estimate overfitting you need to assess your model on an independent dataset. Or use cross-validation, or something similar.