I'm using randomSurvivalForest to predict survival from a rather small
sample. As it's not enough data to have training and validation set, I
rely on the "Estimate of error rate" computed by the randomForest. If I
understand the method correctly, it repeatedly partitions the data into
varying training/validation sets during the learning steps, which also
produces the estimate of error.
My questions is, would it be possible to compute a ROC curve during RF
A possible approach I considered would be to train the RF on a subset of
the data and create a ROC curve from the prediction on the remaining
data. By repeating this process, I would get the variation of the ROC
curve for the different possible data subsets.
But this seems to be not such an elegant solution, as this could be done
during one instance of RF learning.
It would be very helpful, if someone could point to an approach on how
to do this.