A NEURAL NETWORK MODEL TO SOLVE DEA PROBLEMS

Data envelopment analysis (DEA), occasionally called frontier analysis, was first put forward by Charnes, Cooper and Rhodes in 1978 [1]. It is a performance measurement technique which can be used for evaluating the relative efficiency of decision-making units (DMU's) in organizations. The major advantages of DEA are its allowing the relative efficiency to change over time and requiring no prior assumption on the best solution frontier; therefore, lots of businesses or organizations have applied DEA to find their operating performances for further making decisions on the efficiency improvement. Those inefficient DMUs can be identified and proposed to make up their input resources and/or generated benefits. The identification has been performed widely by linear programming (LP) technique [2,3,4]; yet, the serious dependence on the number of DMUs causes the LP technique to a longer DEA computation time. To overcome this limitation, an alternative approach seems to be needed. DEA for a large dataset with many inputs/outputs would require huge computer resources in terms of memory and CPU time. This paper proposes a neural network Data Envelopment Analysis to address