After I am using the Decision tree operatior to "split" my data. I would like to have a label, in which "case" the example is grouped instead of a true/false prediction. Somehow I would like to have am mixture of clustering adn decision tree (Cluster by decision).
DecisionTrees assigns the same confidence for all instances fallen into the same leaf. If you assume that this confidence is different for all pairwise different leaves, you could use Numerical2Polynomial and ChangeAttributeRole or something like this to generate a clusterattribute.
correct me if I got you wrong. You are suggesting using a decision tree as clusterer by assigning all examples of a node into one cluster?
Hmm, this would result in an orthogonal separated subhypercube of the complete data space. Seems to be interessing, I will add this onto my todo list. But I'm not too optimistic that it will be implemented in near future, since some of these interesting ideas already got very familiar with my toDo list...
The way around steffen described might indeed give the desired results. But only if there aren't two leaf nodes with the same class distribution. So the result might be very unstable.