Hi folks!

I am training a set of different models where my goal is to optimize the typical scores like F1, but I am also interested in getting a model ensuring a near zero overfitting. Chatting with ChatGPT :P, it suggested me to implement a custom scoring function where in addition I could balcance two metrics like F1 and overfitting gap during the optimization process.

What do you think about that?
How did you deal with this problem in the past?
Is there something implemented in Scikit-Learn or other library for this purpose?