Rule Induction in Cascade Model Based on Sum of Squares Decomposition

Abstract

AbstractA cascade model is a rule induction methodology using levelwise expansion of an itemset lattice, where the explanatory power of a rule set and its constituent rules are quantitatively expressed. The sum of squares for a categorical variable has been decomposed to within-group and between-group sum of squares, where the latter provides a good representation of the power concept in a cascade model. Using the model, we can readily derive discrimination and characteristic rules that explain as much of the sum of squares as possible. Plural rule sets are derived from the core to the outskirts of knowledge. The sum of squares criterion can be applied in any rule induction system. The cascade model was implemented as DISCAS. Its algorithms are shown and an applied example is provided for illustration purposes.