I was impressed with While's answer here: Relative importance of a set of predictors in a random forests classification in R
I was wondering if anyone has found a package or algorithm to group variables for RF Importance not for Gini in RandomForest, but for the permutation method (type = 1) example code: importance(rf.obj, type=1) vs. importance(rf.obj, type=2). I'm using While's algorithm, but wonder if it applies similarly.
Additionally, the full variable list shows importances that do not change when including grouped variable sets. i.e., Grouping seems to be post-hoc. However, according to research, grouped variables should change the importance metrics for the remaining singular variables. The algo doesn't achieve this, but I'm sure I'm missing something.
