Discriminate approach for data selection in data envelopment analysis

DEA (Data Envelopment Analysis) is a well-known method for evaluating management efficiency of DMUs (Decision Making Units). To calculate efficiency of DMUs, analytical data are necessary. However, there are not clear criteria for data selection so that analysts have to choose the data on their own. Therefore, it is important to support data selection by reasonable ways to let analysis be informative and beneficial. In order to deal with this matter, new methods are proposed based on traditional ones. Support for data selection is realized by considering analyst's intention. Analytical data for making some specific DMUs efficient are obtained by reflecting knowledge or experience analysts have. TDS-DEA (Tight Data Selection based DEA) reflects the analyst's intention strongly and tries to make only intended DMUs efficient. On the other hand, LDS-DEA (Loose Data Selection based DEA) reflects it loosely and at least intended DMUs can be efficient. Then both methods should be examined more detail and how data selection is carried out effectively. On this point, this study prepares the experimental data to clarify the effectiveness and drawback of the methods. According to the experimental result, additional ideas such as discriminate approach or assurance region method are considered to improve the quality of data selection.