Relevancy in constraint-based subgroup discovery

This chapter investigates subgroup discovery as a task of constraint-based mining of local patterns, aimed at describing groups of individuals with unusual distributional characteristics with respect to the property of interest. The chapter provides a novel interpretation of relevancy constraints and their use for feature filtering, introduces relevancy-based mechanisms for handling unknown values in the examples, and discusses the concept of relevancy as an approach to avoiding overfitting in subgroup discovery. The proposed approach to constraint-based subgroup mining, using the SD algorithm, was successfully applied to gene expression data analysis in functional genomics.