I have a imbalanced data, the binary logistic model performs poorly on train dataset with an AUC of 0.66 but it performs well on my test data with an AUC of 0.84. Is it possible and can the model be relied for future purpose. Can covariate shift be a reason? But after doing SMOTE both train and test was similar but sensitivity of test was just 0.53. How do i proceed further?
There might be a high possibility that your test file contains duplicate records of that in train set.
Please check for this possibility and reply with the new results.