Creating new Adaptive Model (ADM) vs Adding new predictors into existing one

General
, , ,
1

We have matured Adaptive model (ADM) on production environment with 70 predictors. Now as per the new requirement we need to consider 10 more predictors conditionally, it means new predictors will have values only in a specific scenarios.

My questions is:

  1. Should we include new predictors into the existing Adaptive model (ADM) ?
  2. Or should we create new Adaptive model (ADM) by coping existing model rule/learning and include new predictors? and update BAU strategy to trigger new model in a specific scenario, and for rest of the scenario trigger the existing model.

As per my understanding, if we go with the option#1, then the new predictors will be inactivated in BAU scenario since it will not have any values.

However option# 2 will add extra overhead to maintain the new model, providing the feedback and to maintain two models instead of one.

Please suggest, which would be better approach to proceed ?

Thank you in advance.

2

@AbhinayU I think you gave a good summary of the possibilities.

It really sounds like ideally you would have two different models and have different sets of predictors for each. However splitting up a model and carrying over the “learnings” is not a supported operation. It would first have to acquire data - maybe in a sort of shadow mode while still using the old one for scoring.

Depending how much of the data will be missing it may not be a problem to stick to just one model. There will be a separate bin for the missing data. Only if the missing percentage is very high the predictor will become inactive.

-Otto

3

Thank you @Otto_Perdeck for your quick response and valuable suggestion, I agree with suggested approach.