1

I have a dataset which has double 0' than 1'. I apply a logistic regression and a logistic regression with weights that balance the train dataset. My problem is when I create ROC curves in both cases they remain the same and AUC is the same.Or better it changes so little that difficult to distinguish. The accuracy of course decreases. I use python and the roc_curve() where I provide the response variable of the test dataset and the probabilities that the model calculates. I have the same behaviour with Average Precision Curve.

I have difficulties to justify it.

Can someone please explain to me?

gbarel
  • 11
  • 2
  • Possible duplicate of [AUC and class imbalance in training/test dataset](https://stats.stackexchange.com/questions/260164/auc-and-class-imbalance-in-training-test-dataset) – Sycorax Jul 30 '18 at 02:57
  • 1
    Why would you expect it to change? When you use weights, you affect the cut point, not the curve. – Peter Flom Jul 30 '18 at 11:32

0 Answers0