Hi Valeriy,

Yes, I think that's an issue. It shows promising results according to AUC, but the probabilities do not look good.

It's on my research todo list :)

My first attempt will be to couple this with something like isotonic regression or platt scaling