A NEW ANN TRAINING APPROACH FOR EFFICIENCY EVALUATION

In this study, we propose a new Artificial Neural Networks (ANN) training approach that closes the gap between ANN and Data Envelopment Analysis (DEA), and has the advantage of giving similar results to DEA and being easier to compute. Our method is based on extreme point selection in a bandwidth while determining the training set, and it gives better results than the traditional ANN approach. The proposed approach is tested on simulated data sets with different functional forms, sizes, and efficiency distributions. Results show that the proposed ANN approach produces better results in a large number of cases when compared to DEA.