Good ROC curve but bad recall curve

Question

I have some machine learning results that I don’t quite understand. I am using python sciki-learn, with 2 million data, about 14 functions. The “ab” classification looks pretty bad on the critical curve, but the ROC for Ab looks as good as the classification of most other groups. What can explain this?

Answer 1

Class imbalance.

Unlike the ROC curve, the PR curves are very sensitive to imbalance. If you optimize your classifier for a good AUC on unbalanced data, you are likely to get poor results accurate to a recall.