I am training a BDT with PCA applied to the input variables and have a question regarding the variable ranking that is provided by TMVA after the training: Are the variables given in the ranking the original variables or are they referring to the principal components after transformation?
The variable ranking is obtained by obtained looking on how much the variable are used when splitting the decision tree nodes. It is then referred to the input given to the trees, so if you apply some combinations of variables, will be referred to that combinations.