Akram Oveysiomran
Abstract
Input and output selection in Data Envelopment Analysis (DEA) has many important. In this research, inputs and outputs of reginal power companies are selected with artifitial neural network. The application of neural network in the selection of inputs and outputs of reginal power companies is not a precedent ...
Read More
Input and output selection in Data Envelopment Analysis (DEA) has many important. In this research, inputs and outputs of reginal power companies are selected with artifitial neural network. The application of neural network in the selection of inputs and outputs of reginal power companies is not a precedent in the literature and it is considered the main advantage of the proposed method. In order to train two layers MLP neural network, after presenting of error resilience, learning method was used. After neural network training, neural network performance is examined by using the test set. RMSE value for 15 test set equals 0/0269 which reflects the high accuracy of training network. The Sensitivity Analysis of the studied parameters which are the same inputs and outputs of Data Envelopment Analysis, with ten percent increase of parameter, compared to the prior one was carried out and output relative error average for neural network parameters was calculated. Based on the output relative error average, inputs and outputs were determined. By comparing the efficiency scores of regional electricity companies before and after reducing the number of variables, it is noticed that the number of efficient companies during the above four periods decreased from 50 percent to 11 percent. Finally, the neural network application in inputs and outputs selection of the regional electricity companies was unprecedented in the literature and this is the main advantage of this method.
