Has anyone (is it possible) to use a decision tree for a multi-level target (similar to multinomial regression)? There is some documentation in SAS on multiple targets, but I think it is referring to totally different targets vs. a single target with multiple values.
Is there an option more suitable than decision trees for this? I've always dealt in world of dichotomies (0,1) (Y/N) etc. Thanks.
EM decision tree does handle a single target column with multiple distinct values. See the basic sampsio.dmairis model for example.
Only the neural network directly handles multiple concurrent target columns in one objective function. In fact, the targets can be of mixed types, for instance multiple each of interval, ordinal, and nominal.
The two stage node and the group processing nodes enable modeling multiple target columns sequentially.