I have a relatively small dataset n = 500 for which I am training a CART decision tree. My dataset has about 30 variables and the outcome has 3 classes. enter image description here

I am using CART for interpretability purposes, as what I am interested in, is sharing and analyzing the rules generated by CART.

However, I don't know what would be the optimal way to externally validate my CART model while still maintaining the intrinsic interpretability of CART.

The only option I came up with was using 80/20 split which I find relatively weak. Intuitively, I would like to use bootstrapping which I think would be more robust in terms of justifying the lack of external dataset but I don't think bootstrapping would translate in CART.

The question is open. How would you approach validation (other than the generic 20/80 split) for a CART decision tree model?

As of note, this task is for a potential publication in a medical journal.

Thank you!



Your Answer

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