Subgroup Discovery with Consecutive Erosion on Discontinuous Intervals

The subgroup discovery problem aims to identify a subset of objects which exhibit interesting characteristics according to a quality measure defined on a target attribute. In this paper, we propose a new optimized approach, called SD-CEDI, which originality consists of extracting subgroups defined on discontinued attribute intervals. The intuition behind this approach is that disjoint intervals allow refining the definition of subgroups and therefore the quality of the subgroups identified. For this purpose, the approach we propose models the search space through a hypercube and aims to slice this hypercube to highlight subgroups. Unlike recent methods that also exploit the concept of hypercubes, the originality of our approach relies on the way it performs the slicing in order to bring out discontinuous attribute intervals. The good performances of the approach are demonstrated by comparing, qualitatively and quantitatively, the results with the main algorithms that are the references in the domain.