This paper is concerned with the induction of classification trees for unbalanced data, i.e. for the case where some categories of the target variable are much less frequent than other ones. More specifically, we address two aspects. On the one hand, we look for growing criteria that efficiently take into account the specific unbalanced nature of the data. On the other hand, we deal with the relevance of the conclusion that should be assigned to the leaves of a grown tree. We have recently considered two independent ways for dealing with these issues. The first one consisted in defining and using out centered entropies, and the second one on relying on measures of implication strength derived from implicative statistics. The aim of this paper is to compare and establish the relationship between these two approaches.