Bayesian based subgroup discovery

Abstract

Data Mining is concerned with extraction of interesting patterns or knowledge from huge amounts of Data. Generally data mining tasks are either predictive or descriptive. Classification falls under predictive induction while clustering and association rule mining fall under descriptive induction. Subgroup discovery is a task at the intersection of supervised learning and descriptive induction. In subgroup discovery we want to uncover individual patterns in data with a given property of interest. We want to find subgroups that cover a large population and are statistically different. The main application areas of subgroup discovery are exploration and descriptive induction, where the user wants to find the overview of dependencies between a target and many explaining variables. Many techniques have been proposed for discovering subgroups and some of these techniques are based on classification. But none of the techniques uses Bayesian networks for the generation of subgroups. Our contributions include a technique for the discovery of subgroups where the subgroups are generated using Bayesian networks.