Let's consider an e-commerce problem.
I have data about users that almost place an order online : they give some information about themselves (name, age, etc.), but won't immediately validate the order.
Sometimes they'll come back to finalize it, but other times, they'll never be back (better offer/product at another place). I do have the data to know if a user came back or not, and if he came back, how long it took him to come back.
Thus, I am interested by 2 points :
- Will the user come back ?
- If yes, how long did it take him to do so ?
For the first problem, it is obviously a classification problem. And I am considering a regression problem to answer the second question as it is a duration problem.
However, I was wondering if I should train these models separately, or if a specific method to take into account the link between these two was more appropriate ?