Any cut point divides the domain of one attribute in two partitions having
approximately the same class distribution as the entire domain. Thus a
condition on a single attribute does not improve class separation, so univariate
supervised discretization methods based on information class entropy are
unlikely to work.