3
$\begingroup$

There are multiple interpretations of area under ROC curve. (e.g What does AUC stand for and what is it? ). We also know that AUC is closely related to rank correlation. Are there also different ways how to visualize it?

I am not thinking about the whole curve, but about AUC as a measure. Maybe some scatter plots or histograms that shows the rank correlation, or something related to proportion of pairs of samples with higher predicted probability for positive sample than negative.

$\endgroup$
  • 2
    $\begingroup$ The AUC is a scalar (unitary) value. Anything that represents a single number would do it. Would you be interested in different ways to represent the ROC curve? $\endgroup$ – gung Dec 13 '16 at 17:40
  • $\begingroup$ Related: stats.stackexchange.com/questions/145566/… $\endgroup$ – Sycorax Jul 30 '18 at 4:16
2
$\begingroup$

AUC can be visualized with a scatter plot. Suppose the points in the plot correspond to such pairs of data points in the evaluation set that one of each pair is a negative example n and the other one is a positive example p. If the x coordinate of a point is the predicted probability of n belonging to the positive class and y coordinate is the predicted probability of p belonging to the positive class, then the proportion of data points above the diagonal equals AUC.

You can check an example in this blog post.

$\endgroup$

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.