Kenneth W. Flowers

Beth A. Novick

Douglas R. Shier

Abstract: This paper considers the size assortment problem, in which a large number of size distributions (e.g., for retail stores) need to be aggregated into a relatively small number of groups in an optimal fashion. All stores within a group are then allocated merchandise according to their common size distribution. A neighborhood search heuristic is developed to produce near-optimal solutions. We investigate the use of both random and "intelligent'' starting solutions to initiate the heuristic. The intelligent starting solutions are based on efficiently solving one-dimensional versions of the original problem and then combining these into consensus solutions. Computational results are reported for some small specially structured test problems, as well as some large test problems obtained from an industrial client.

Key Words: clustering, consensus, heuristic, matching, minimum clique, neighborhood search