Hi,
I'd appreciate your thoughts about the effect of highly linear dependencies
among predictors on Boosting Trees. My thinking is the following: if the
objective is prediction, then multicollinearity isn't a problem. Now, if the
objective is also interpretation, I'm wondering how, for instance, partial
dependence plots might be affected by correlated predictors. In a single
Tree, if say two predictors are correlated then only one might be used in
the model, but in a Boosting tree model I have a collection of trees, so a
given predictor has more chance to make it into the model. Will partial
dependence calculations be affected by correlated predictors?
Thanks in advance,
Lars.
You may leave the list at any time by sending the command
SIGNOFF allstat
to [log in to unmask], leaving the subject line blank.
|