Using SAS to find the most important variables out of a large number of variables

[de95]

Hello All,

I have a dataset with over 500k observations and about 500 variables. About 20 variables are categorical, nominal, or datetime and the rest are numeric. I want to build a model using this dataset. I have a dependent variable that is binary (1 or 0) but there are too many dependent variables. I want to reduce the number of dependent variables to about 30. Someone suggested I use a random forest and an importance plot to find the 30 most importance variables. I have never used random forest before but I have a basic understanding of the theory behind it.

edit: Also someone suggested Chi-square for feature selection.

Could you please show me an efficient way to find the 30 most important variables. I am using SAS enterprise 7.1.

Any help would be great

[Ksharp]

I would recommend to use PROC HPGENSELECT .

or

PROC PLS + missing=em  option (which could better handle missing value).

If your variable have many missing value ,try PROC PLS .(HPGENSELECT would drop these missing obs)