decision tree interpretation, variable importance of the split as predictors?

Dublin187 — Fri, 15 Mar 2019 17:26:44 GMT

Hi all,

I have built a decision tree in SAS VIsual Analytics and I am not sure how to interpret it despite all the tutorials and readings I did.

As I understood, the Variable importance plot on the right of the screen displays the predictors in order of importance. (the one with the highest score and darker fill will the most important and so on, )

Also, as I have read, the decision tree plot displays the first split as the most likely predictor, the second split will show the second most likely predictor ...

Did I understand that correctly?

My problem is the following : when I analyse the decision tree, the order of the importance of the variable does not match the order of splits (the most likely predictors of the decision tree. )

Why are the two not showing the same order?

What should I conclude from the decision tree?

I am unable to draw some conclusion because I don't know what to prioritize in my interpretation of the who decision tree model.

Your help on the matter would be really appreciated.

Thank you

Re: decision tree interpretation, variable importance of the split as predictors?

Carlos_Lara — Fri, 20 Mar 2020 14:10:50 GMT

Here is the explanation for why this happens

https://go.documentation.sas.com/?docsetId=casml&docsetTarget=viyaml_treesplit_details20.htm&docsetVersion=3.0&locale=en

topic Re: decision tree interpretation, variable importance of the split as predictors? in SAS Visual Analytics

decision tree interpretation, variable importance of the split as predictors?

Re: decision tree interpretation, variable importance of the split as predictors?