Inductive Querying for Discovering Subgroups and Clusters

We introduce the problem of cluster-grouping and show that it integrates several important data mining tasks, i.e. subgroup discovery, mining correlated patterns and aspects from clustering. The problem of cluster-grouping can be regarded as a new type of inductive optimization query that asks for the k best patterns according to a convex criterion. The algorithm CG for solving cluster-grouping problems is presented and the underlying mechanisms are discussed. The approach is experimentally evaluated on a number of real-life data sets. The results indicate that the algorithm improves upon the subgroup discovery algorithm CN2-WRAcc and is competitive with the clustering algorithm CobWeb.