Feature Importance

dK00

Hello Everyone,

I would like to know if I can extract the feature importance out of gradient boosted trees model in Rapid miner. I have tried connecting the weight output, but the weight is in thousands for each variable. Is this method correct? if not, what is the right operator/connection to be used?

Caperez

Hi @dK00, 

It's possible to see the variable importances into the model. 
If you connect the mod port to a res port, for example, into the Boosted tree model, you can go to the description tab and search the section "Variable Importances" There are the variables relative imporances (weights), also relative or scaled.

Best, 

Cesar 

dK00

Hi @ceaperez,

Thank you for your answer, I really appreciate it.

Another question if you don't mind. I am using Boosted tree model for a medical binary classification problem. It is providing a very good performance in comparison to Random Forest and Decision Tree.

Do you think it is the right model to be used? and do you think the feature importance of GBT makes sense and interpretable? 

Thank you in advance!

Caperez

Hi @dK00, 
You welcome.

The sense, the interpretation and the right model depends on the context, your dataset and other factors, it's not a simple question with a simple answer, this is the job of the data scientist,

Best, 

dK00

Hello @ceaperez,

Thank you for your support. I was able to output the variable importance generated by the model. 

I was trying to illustrate the attribute weight as a graph. However, they were showing in thousands. Is there a way to illustrate the scaled values? 

Below is a screen shot of the result.


Thanks in advance.

Caperez

Hi @dK00,

Nice to see you again. Have you tryed the Normalize operator for that?

Best, 

Cesar