Solved – Out of Bag Prediction Error Estimate from Random Forest Regression (i.e., not classification)

I would like to use the OOB cases from a random forest fit to estimate the mean squared prediction error so I don't have to cross-validate. I am using the randomForest package in R. It is clear from the documentation that OOB error is reported for classification, but I can't figure out how to get OOB MSPE for regression. Am I missing it or is it truly not reported, which seems odd?

It's not reported.

But you can use the predicted part of the fit, which as per the help page yields

the predicted values of the input data based on out-of-bag samples.

An example:

set.seed(131) ozone.rf <- randomForest(Ozone ~ ., data=airquality,na.action=na.omit) mean((ozone.rf$predicted-airquality$Ozone[as.numeric(names(ozone.rf$predicted))])^2) [1] 316.7915 

Similar Posts:

Rate this post

Leave a Comment