R help - May 2011 - Random Forest

I'm analyzing data using Random Forest Regression.  For some of the
species I am analyzing, the percent variation explained is negative.
Could you please explain to me what that means?  If you need more
information, please let me know.  Thank you.

 

Sincerely, 

 

Rachel Unger


	[[alternative HTML version deleted]]

On Tue, May 24, 2011 at 3:18 PM, Unger, Rachel
<rachel.unger at cahnrs.wsu.edu> wrote:
> I'm analyzing data using Random Forest Regression. ?For some of the
> species I am analyzing, the percent variation explained is negative.
> Could you please explain to me what that means? ?If you need more
> information, please let me know. ?Thank you.
If you're using the rsq component of the randomForest output, negative
value means the mean error is larger than the variance of the response
y. This could be because the predictor performs really poorly but also
because of some calibration issue.

You may want to consult someone face to face to diagnose the problem.
As a start, I would look at a scatterplot of observed vs. predicted
values to see whether there is a systematic effect

HTH,

Peter