Classification trees for ordinal variables (Q626218)

Classification trees growing algorithms are considered for the case when there is an ordering of the response variable values. New versions of the Gini-Simpson and Twoing criteria are proposed for the choice of the nodes to split, which are consistent with the ordering. It is shown that the obtained version of the Gini-Simpson criterion does not pose the (ordinal) exclusivity preference property. A new criterion is proposed possessing this.