# Recursive Feature Elimination Fails to Output as Expected

Currently I am using rfe function in the "caret" package to do feature selection. There are 380 variables as input candidates. I have done many trials and I noticed that something weird always happens, as the rfe tends to output all candidate variables as the best subset while I have set the parameter size to be 5 to 10.

I am wondering whether anyone else has come across this problem. Any help will be appreciated. Thanks in advance for any help.

• Some code that reproduces the problem (perhaps on a simulated dataset) would be extremely helpful. – Zach Oct 22 '13 at 13:41
• I agree: we can't help unless you are more specific about the code, the data or anything else. Also, RFE always evaluates the largest model even if sizes does not include that value. If the full model is best, that is the outcome from RFE. – topepo Oct 22 '13 at 19:24