2
$\begingroup$

Considering the definition of AUC (Area Under Curve), is that a reliable performance metric for a multi-class (30-40 classes) classification problem?

$\endgroup$
6
  • $\begingroup$ AUC is a binary classification metric, I do not think you can use it for multi-class classification $\endgroup$ Apr 12, 2019 at 10:04
  • $\begingroup$ @RobinNicole You are right but it also has been suggested for multi-class classification by binarizing techniques. $\endgroup$ Apr 12, 2019 at 10:15
  • $\begingroup$ What do you mean by binarizing technics ? If you have a link I am really interested :) $\endgroup$ Apr 12, 2019 at 10:20
  • $\begingroup$ @RobinNicole I think reading this article can be helpful for you: scikit-learn.org/stable/auto_examples/model_selection/… $\endgroup$ Apr 12, 2019 at 11:36
  • $\begingroup$ Which technique do you use for calculation of TPR and FPR? micro-averaging or macro-averaging? $\endgroup$
    – pythinker
    Apr 12, 2019 at 14:04

1 Answer 1

0
$\begingroup$

No - AUC (Area Under Curve) can not be used directly to assess the performance of multi-class classification.

If you want to use AUC, it is necessary to binarize the output. Either each class as to be compared against each other class or 1 class versus the rest.

$\endgroup$

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