topic Re: random forest in SAS Data Science

random forest

moshe_ke — Fri, 27 May 2016 14:27:48 GMT

Hi,

Is it possible to know which variables are most important at Random Forest and how they are selected ?

thanks

moshe

Re: random forest

Ksharp — Sat, 28 May 2016 06:09:27 GMT

Use Decision Tree PROC HPSPLIT . and Check the documentation, there is already an example to do this .

Example 61.5: Assessing Variable Importance


proc hpsplit data=MBE_Data maxdepth=6;
class Usable Dopant;
model Usable = gTemp aTemp Rot Dopant;
prune none;
run;

Re: random forest

moshe_ke — Sun, 29 May 2016 05:25:51 GMT

Thanks ,

Is there a mathematical description of how the explanatory variables selected,

Moshe

Re: random forest

Ksharp — Mon, 30 May 2016 00:46:37 GMT

Yes. Check documentation. Like entropy .....

Re: random forest

JasonXin — Wed, 01 Jun 2016 13:47:16 GMT

Hi, Here is sample code to turn on VI in HPFOREST (please see attached .sas code) There is a section "Measure variable importance" that covers the details on the subject including all the math details. VI is sensitive to split method selected. Hope this helps. Jason Xin

Re: random forest

JasonXin — Tue, 31 May 2016 23:58:30 GMT

Sorry, my previous post had the format of the sample code messed up. Attached is the sample .sas program for the variable importance option. Thanks. Jason Xin