randomforesterror

[Mr96]

Hi, I'm running a regression on the insurance data. I'm doing random forest and I get the following error:
"Warning message:
In mean.default((ytest - pred.bagging)^2) :
argument is not numeric or logical: returning NA"

my code is
library(randomForest)
set.seed(888888)
xtrain<-M[1:1000,]
ytest<-M[1001:1338,]
fit.bagging<-randomForest(charges~.,data=xtrain,importance=TRUE)
fit.bagging

pred.bagging<-predict(fit.bagging,data=(-xtrain))
MSE.bagging<-mean((ytest-pred.bagging)^2)
MSE.bagging

Can anyone help me with this error?

[FJCC]

I haven't used the randomForest package so I am working from general experience.
Shouldn't this line

pred.bagging<-predict(fit.bagging,data=(-xtrain))

be

pred.bagging <- predict(fit.bagging,data=(ytest))

That is, do the prediction with the data that you held back. Also, I would expect that the parameter for passing the data in predict() to be called newdata not data but I could easily be wrong about that.
In the line

MSE.bagging <- mean((ytest-pred.bagging)^2)

you are subtracting the pred.bagging values from an entire data.frame, ytest. Shouldn't you be referring to the predicted values, charges, from ytest?

MSE.bagging <- mean((ytest$charges-pred.bagging)^2)

[Mr96]

would you like to be my friend? if yes, let me know please. I believe I owe you a lot. you saved me again.
Thank you for your time and help and kindness FJCC :* love u
Sorry to bother u again, but do you think the code would be the same for classification too?

[FJCC]

I am not sure of the changes needed for classification. I would inspect the output of predict() to make sure you get the predicted class. I have a vague memory of having to change an argument in predict() to get that but I was not using randomForrest at the time. Also, MSE is not appropriate for describing a classification problem.

[Mr96]

gottcha. thank you for all your help