Combining data envelopment analysis models and machine learning algorithms for evaluating the efficiency of decision-making units considering undesirable outputs

Undesirable outputs are an integral part of production in various decision-making units, and to bring analyses closer to the real world, it is necessary to consider, undesirable outputs in performance evaluation research. In this paper, a new hybrid model for evaluating the efficiency of decision-making units in the oil industry is presented, which uses slack-based data envelopment analysis techniques and advanced machine learning algorithms. The proposed model specifically focuses on improving efficiency considering undesirable outputs and conditions of uncertainty. Three machine learning algorithms including artificial neural networks, support vector machines, and XGBoost are used to predict and improve the results of slack-based models. This study involves the evaluation of 37 decision-making units within the National Petroleum Products Distribution Company, and the results show a significant improvement in efficiency using predicted data compared to actual data. This research not only contributes to new perspectives in efficiency evaluation and improvement but also offers innovative hybrid methods to address challenges in operational management.