One of the problems with data envelopment analysis (DEA) is that it results in too many decision making units (DMUs) as efficient. This leads to a problem of discrimination among the efficient units. Model misspecification and unrestricted weight flexibility are two main reasons for the discrimination problem. In this paper, we propose and test a model averaging ensemble approach that results in unique DMU rankings. We also prove that ensemble based ranking of DMUs will always result in equal or fewer efficient DMUs than any other single DEA model considered in the ensemble.