On the Selection of Meaningful Association Rules

Abstract

Association rule mining produces a large amount of rules. Many of them are redundant ones. This chapter has presented techniques to select rules that are semantically useful and carry the most information. They aim for a complete understanding, rather than an overall picture, about the domain. Prior to the analysis, variables are grouped into subjects which correspond to the domain's perspectives. These subjects are key factors to classify the rules into strongly meaningless, weakly meaningless, partially meaningful, and meaningful ones. Rules that have equivalent patterns are identified and the most significant one is selected. Furthermore, between a general and a more specific rule, the latter is selected since it offers more insight about the domain.