Why it would be too optimistic to compute presicion, recall and f1-score to evaluate a model trained for imbalanced classification on a balanced testing sample ?


1 Answer 1


The test set should represent what your model will encounter in practice when you use it to make predictions. It serves as a resource to demonstrate that what you did will actually work.

In case your test set does not represent what the true problem is or will be in the future, you can‘t say much on how well your model works.

If you test on a balanced set while you have trained on a unbalanced set (which is harder to tackle), your test predictions will likely look better or worse than what you could reasonably expect with the „real“ unbalanced problem. So you cannot say anything with this test about the quality of your model.


Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.