just a question my data is not balanced and to address the imbalance I am using cost sensitive logistic regression however the results are surprising, ROC AUC = 0.69 and PR AUC = 0.007 does the Pr Auc score indicate that my performance is worse than guess? Is working on a binary classification.
Asked
Active
Viewed 79 times
0
-
No that is not how it works for the PR AUC. Generally speaking in scenarios with very high class imbalance it is recommended to use PR AUC over ROC AUC. PR AUC is much more sensitive to improvements with regards to identifying positive samples – Janosch Oct 18 '22 at 14:33
-
You could benefit from a look at https://stats.stackexchange.com/questions/283170/when-is-unbalanced-data-really-a-problem-in-machine-learning – kjetil b halvorsen Oct 18 '22 at 16:41