3
$\begingroup$

Ive tried to understand what the ROC Curve represents and what EER (Equal Error Rate) means. And I somehow think I got to understand some of the explanations I read on the internet and videos I watched and papers I read. But I still cant get a grip on what it means in my particular case.

I am reading a paper on face verification. The described method is trained and tested on the "Labeled Faces in the Wild"-Benchmark. That benchmark offers a standard dataset for comparison to other face-verification-methods. It essentually is a huge database of image-pairs which each depict the same or different persons. The methods at hand should now tell if the depicted persons on these image-pairs are the same or different people. The machine learned method is evaluated with a measure called ROC-EER and the score/result is stated in percentage. And they also evaluated by doing a 10-fold cross validation (which I actually do understand and know what is being done and why)

For example: ROC-EER, %: 89.5

So I just dont know what that 89% should tell me? 89% of what is what? And how does that value correlates to the 10-fold cross validation procedure?

And I even dont know If I am missing some information for you to answer a complete question...I feel a bit lost here :)

Improve this question
$\endgroup$
2
  • $\begingroup$ Please, could you link the paper and/or some explanation of EER? I haven't met the term before, although I could make guesses what the term may mean... $\endgroup$
    – cbeleites
    Commented Jan 28, 2015 at 8:02
  • $\begingroup$ The paper can be found here google.de/… and the explanations in question are on page 6 and the table on page 7 $\endgroup$
    – degill
    Commented Jan 28, 2015 at 8:28

1 Answer 1

Reset to default
3
$\begingroup$

The paper you link explains ROC-EER as

Receiving Operating Characteristic Equal Error Rate (ROC-EER), which is the accuracy at the ROC operating point where the false positive and false negative rates are equal

So the EER is a way to fix the working point (= threshold for deciding the class) for an actual classifier on the ROC.

So the 89% you quote are then the overall accuracy if you fix the threshold according to the ROC-EER method.

Cross validation is the method ("measurement schedule") how they measured the 89%

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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