Interaction as an interestingness measure

Abstract

Detecting interesting patterns in data has been a focus of recent work in knowledge discovery. Understanding the patterns of interaction between attributes is relevant to many fields. Existing measures of interestingness do not adequately detect these interaction patterns. Here we present a new measure that explores the interactions to be found in data. We combine this interestingness measure with statistical validation to find reliable and interesting interactions. We first develop the concepts of interaction in terms of interestingness. We then demonstrate use of interaction rules to find interesting patterns in datasets from diverse domains. © 2010 IEEE.