Tackling Reject Inference

Tackling Reject Inference - by Augmentation or Extrapolation.

a. Augmentation :

i. Build a logistic model (M1) with all booked and non-booked loans for book or non-booked DV.
ii. Score the booked loans by this model M1. Weight the loans by reciprocal of the probability derived by this regression model.
iii. Then build the scoring model for booked loans in weighted development sample.

b. Extrapolation :

i. Build first logistic model (M1) for booked loans on good/bad DV.
ii. Score the rejects and find the probability of good/bad for each. Define DV for rejects based on this pred DV.
iii. Build scoring model (M2) now with both the booked loans and rejects.
(Rejects can be weighted based on estimated bad rate.)

This way tackling the reject inference is expected to help in increasing the approval rate in future.