I'm running a series of regressions using decision trees and am getting good results, but I've got a question.
In the pacakge rpart, you can run cpplot to get a graphical representation of where to cut the tree. As I understand it you choose the split with lowest xerror value and prune the tree at that point.
The issue I'm having is that my graphs are effectively horizontal below the line after the first two splits, but based on my work with the data set over the last several years I think the bigger trees more accurately reflect the relationships.
So my question is this...
Given that the "deviance" is the result of cross-validation, if the overall RMSE is smaller for the larger trees, why is R suggesting a smaller tree in the first place?