ROC analysis and perfcurve

I am trying to perform ROC analysis. The function perfcurve is excellent. However, it assumes that larger values of score indicate stronger evidence for a positive state. This is exactly what you want to do if you are analysing the output of a classifier model.

In my case, I want to use raw data where smaller values indicate stronger evidence for a positive state. Reading through the documentation, I could not find a way to change the test direction for the perfcurve function.

Does anyone know if it is possible to do this? Or other similar functions that allow doing this?

Answer 1

[Ilya]

How about putting a minus sign in front of the scores (2nd input to perfcurve)?

[Costanzo]

Yes, that does the trick. However, ideally I am looking for a solution that doesn't require manipulating the inputs (or the outputs) of the function.

Answer 2

[Muge Karaman]

Hi Costanzo,

I have been having the same issue using perfcurve, and been manipulating the input by multiplying by -1. However, as you said, this is not ideal because the threshold values would then need to be manipulated in the same way if needed to be reported. Please update this post, if you have found a solution to this!

Thanks,