An alternative clustering approach: a DEA-based procedure

The problem of clustering objects into groups is a branch in statistical multivariate analysis. This problem is usually solved using a statistical heuristic or an optimization. This article addresses the issue of investigation of the relative efficiency of operational units while taking into consideration the size of units in the sample. This article proposes a two-step procedure allowing managements, first, to cluster the operational units into different clusters, and second, to evaluate each unit in its cluster. The clustering problem is formulated as a multi-objective linear program and the set of units are partitioned into k groups based on their size. The proposed clustering approach in this article employs common weights derived from a data envelopment analysis-like model to cluster the data with input and output items. The performance of the proposed approach is tested on a real problem.