Feature Importance for Regression Random Forest

omohsen

Hello Everyone,

I am looking for an operator (or any other way) to find the attribute importance of my model.
I have selected an RF model and tried to use the operator "Weight by Tree Importance" to find the weights of my attributes. However, I received the following error message:
Attribute Weights cannot be extracted from regression trees.

My dataset contains numerical and nominal attributes, and the label is numerical (double).

rjones13

Hi @omohsen

I would recommend having a look at the interpretations extension, see: https://community.rapidminer.com/discussion/59025/new-extension-interpretations-shap-lime-and-shapely

Please let me know how you get on.

Best,
Roland

rjones13

It is also worth noting that the random forest operator already produces a weights object which gives you information on the importance of different variables. The extension I mention above just adds more granularity and is appropriate to most models.