ItemSB: Itemsets with Statistically Distinctive Backgrounds Discovered by Evolutionary Method

Abstract

In this paper, we propose a method for discovering combinations of attributes (i.e. itemsets) against a background of statistical characteristics without obtaining frequent itemsets. The method considers a database with numerous attributes and can directly find a combination of highly correlated attributes from small populations in two consecutive variables of interest even from an incomplete database. As the proposed method determines local patterns in large-scale data, it may be used as a basis for large-scale data analysis. Evolutionary computations characterized by a network structure and a strategy to pool solutions are used throughout generations. Moreover, association rules are used to generalize the analysis method as itemsets with statistically distinctive backgrounds (ItemSBs). The class-association rules used for classification constitute a discovery method of attribute combinations, which are characteristic when the ratio of class attributes is obtained. The proposed method is an extension to statistical bivariate analysis. In addition, we determine contrast ItemSBs that are statistically different between two subgroups of data while satisfying the same conditions. Experimental results show the characteristics and effectiveness of the proposed method.