講演題目:Streamlined DEA computation methods for large scale DEA problems
講演概要:As a production efficiency or performance evaluation tool, DEA evaluates each DMU with all DMUs as references conventionally. Thus, computational challenge arises when large number of DMUs exist. However, only DMUs on the efficient frontier play as the benchmarks for evaluation. So, the key to release the computation burden is to reduce the model size by finding out those efficient ones in advance. Inspired by the pre-score scheme which discriminates exterior and interior DMUs respect to a selected subsample, we proposed a faster approach based on 2 lemmas found in our research, which includes four steps briefly: (i) Select a subsample of strong efficient DMUs. (ii) Find the potential benchmarks respect to the mentioned subsample. (iii) Identify the efficient frontier. (iv)Evaluate all DMUs. Meanwhile, we further developed an approximate approach by grouping DMUs corresponding to the ‘possibility’ of being benchmarks. The required running time can be reduced much more significantly with relatively accurate evaluation result by our brand new approximate method.