Surrender model having model base of 1 year data (apr17-mar18) with event rate 13% is holding well in train -test data with AUC of 0.83. Model is not overfit & built with variables with IV range between 0.02 to 0.5, but when i am testing the model on oct-nov-data 2018 ,event rate 3% it is not holding.

i understand model base & validation base should have almost same event rate otherwise model will never hold. Is that correct? If no how can we make this model to perform well on some unseen data?

0 Answers