I need to develop a new credit default classification model for which there are a lot of features available but very few historical data (because it's a new activity launched by the company I work for). Because of the lack of historical data, I am quite sure the model won't be able to extract meaningful insights from training data. However, I have good intuition about what decisions the model should take once it is more mature : like I know it should give better grades to borrowers with more money, with better credit history etc. Then, what do you think would be a good approach :

  1. Implement basic rules until the model becomes more mature, i.e. rules like : "don't lend to borrowers with less than X on their account and with less than Y past default-less loans" etc. But the problem I see with this approach is that the model will then be biased by the rules I have decided, because it will only know about the borrowers that passed the rules.
  2. Let everyone pass and consider that the default cost caused by this approach is the price to pay to have a good credit-default model (not sure I can get that accepted)
  3. A mix of 1 and 2 : if it passes the rules then accept the loan, else let it pass with X% chance (random acceptance). But this approach still creates a bias in the model because it will see more borrowers that passed the rules than there really is out there.

What do you think of above approaches ? Can you think of any other good solution ?


  • $\begingroup$ Look into simulation. $\endgroup$
    – GooJ
    Commented Sep 19, 2022 at 10:11
  • $\begingroup$ Hey @GooJ would you please be able to give an example or send some a link that describes what you propose ? Thanks $\endgroup$
    – Anatole
    Commented Sep 19, 2022 at 10:12


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.