Random Forest for Variable Selection

[EricTsai]

Hi,

Does anyone know how to use random forest in Enterprise Miner to do variable selection?  I have large number of original variables to start with ranging from 7,000 to 8,000.

I have tried using Variable Clustering Node in EMiner but would love to learn more ways of variable selection for large data set.

Thanks a lot.

[M_Maldonado]

Hi Yi-Chun,

By default your HPForest node has the property Variable Selection set to Yes. All you need to do is run an HPForest node, and then connect any other node. The the variables that have an out-of-bag reduction less than or equal to zero get rejected after your HPForest.

good luck!

miguel

[EricTsai]

Hi, Miguel:

In your experience, if you have thousands of potential independent variables to start with in your modeling process, what would you try to reduce this huge number to a more manageable list?

Have you seen any method that give you a more predictive model at the end?

What is the state of art in this area right now at SAS?