0
$\begingroup$

I have used two regression models from ML for a specific problem. Now I want to study some features importance using the the partial dependence plot "PDP". My question is: do I need to obtain a PDP for each regression model? or I will get similar plots?

Thanks

Improve this question
$\endgroup$
1
  • $\begingroup$ Please explain what you mean by the acronym. $\endgroup$
    – Tim ♦
    Oct 8, 2021 at 11:09

1 Answer 1

Reset to default
1
$\begingroup$

I'm assuming by PDP you refer to "Partial dependence plots". If so:

  1. You should expect to get straight lines as the models are linear
  2. If the models contain different variables then you should expect to get different line slopes for the same variables
  3. PDP is usually not the best tool for feature importance study. In the case of linear regression you would probably want to either look at the variable coefficient P-value or measure the reduction in accuracy measure associated with the removal of each variable. You can read some more about feature importance measures here
Improve this answer
$\endgroup$
1
  • $\begingroup$ Yes indeed, I have used ANN, DNN, SVR, RF to solve two nonlinear regression problems. and I have added a new feature (the same) for each problem, the accuracy for the first problem gets better, where for the second problem it does not change. So I thought to used PDP to study the dependency plot of this feature for both problem to try to understand its importance. $\endgroup$
    – jojo
    Oct 8, 2021 at 10:46

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.