The "type=1" importance measure in RF compares the prediction error of each tree on the OOB data with the prediction error of the same tree on the OOB data with the values of one variable randomly shuffled. If the variable has no predictive power, then the two should be very close, and there's 50% chance that the difference is negative. If the variable is "important", then shuffling the values should significantly degrade the prediction in the form of increased MSE. The importance measure takes mean of the differences of all these individual tree MSEs and then divide by the SD of these differences.
With that, I hope it's clear that only v2 and v4 in your example are potentially "important". Best, Andy -----Original Message----- From: r-help-boun...@r-project.org [mailto:r-help-boun...@r-project.org] On Behalf Of Johnathan Mercer Sent: Monday, August 27, 2012 11:40 AM To: r-h...@stat.math.ethz.ch Subject: [R] interpret the importance output? > importance(rfor.pdp11_t25.comb1,type=1) %IncMSE v1 -0.28956401263 v2 1.92865561147 v3 -0.63443929130 v4 1.58949137047 v5 0.03190940065 I wasn't entirely confident with interpreting these results based on the documentation. Could you please interpret? [[alternative HTML version deleted]] ______________________________________________ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code. Notice: This e-mail message, together with any attachme...{{dropped:11}} ______________________________________________ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
