I made an ROC curve for three classifiers and am getting the following results:
My blue and orange lines are going under the diagonal and I'm pretty sure that's not supposed to happen. Does anybody have an explanation for this?
Asked
Active
Viewed 490 times
1
-
Yes, it's a bad model. – Denziloe Aug 23 '18 at 09:37
-
Yes the response changes. First you predict the opposite then you are ok so you probably have a non linear response. The Forest missclassifies at the beginning, wrong buckets size I guess. Green just gets away with it, but you have mix responses at the beginning for lower value. Your data is probably mixed up there with 0 and 1 difficult to separate and then beyond a threshold it becomes easier. Then depends on sample size... – R. Prost Aug 24 '18 at 05:42