Benchmarking by an Integrated Data Envelopment Analysis-Artificial Neural Network Algorithm

Data envelopment analysis (DEA) is a nonparametric approach using mathematical models, which evaluates the efficiency in a set of decision making units (DMUs) and offers the benchmarks to the inefficient units to better performance. Artificial neural networks (ANNs) are configured for specific applications, such as pattern recognition, function approximation, data classification and so on in different areas of sciences. In this paper an algorithm is proposed using DEA and ANN for efficiency analysis and benchmarking. One of the important issues, from the managers’ point of view, is to improve the efficiency of the DMUs by altering a given parameter and subsequently finding appropriate benchmark for this DMU. In the four-stage proposed algorithm, first the efficient units are identified by DEA, then the coordination of inputs and outputs related to the efficient DMUs are used for training the ANN in order to establish a correlation among these entities. Managers’ desired inputs are given to the trained ANN, so the outputs are estimated for future. The new set of input-output coordination is applied to DEA in order to analyze the performance and obtaining benchmark to the inefficient DMUs. The proposed algorithm has been incorporated in a banking system. The results of this algorithm provides useful information on the evaluation of DMUs’ efficiency and also benchmarking for the inefficient DMUs, for future periods based on the managers’ desired input values