Predicting an event that is conditioned to another

Hi, I want to predict an event let’s say: “a client not paying the bill of his credit card”, however, the problem begins with not knowing if this client is going to activate or not his credit card (some request a credit card but never use it).

What suggestions would you give? Should I predict the probability of activation as well?
Any answer will be highly appreciated. Thanks.

First of all, it would depend on what business goal are you trying to achieve?
Think about this (one of many) approach.

Designing analytics backwards: decision < conclusions < analysis < data < execute

  1. What decision should be made?
  2. What conclusions will help to make a decision?
  3. What analysis will provide the necessary conclusions?
  4. What data is needed for such an analysis?
  5. Execute

Now, card activation status could be one of the predictors for bill payments, agree? So in general, this sounds like two different tasks - predict card activation and predict bill payments.

For the second one you may want to filter your dataset and process only those, who have their cards activated.

1 Like

Hi, thanks for your quick reply, I’m trying to set a threshold to accept/deny a credit card, but my model -built with only activated cards- is overestimating the overall risk.
I’m taking your advice and estimating a prediction for the card activation as well.

1 Like