The reason of poor performance of decision tree
classifiers in minority class is that most of the classifiers employ a post-pruning method. Any node can be
removed and assigned the most common class of the
training instances that are sorted to the node in question.
Thus, if a class is rare, decision tree algorithms often
prune the tree down to a single node that classifies all
instances as members of the common class leading to
poor accuracy on the instances of minority class.