Integrating the neural network into the stochastic DEA model

Abstract

The novelty of integrating neural networks (NNs) into the stochastic DEA model lies in the ability to address some of the limitations of traditional DEA models and improve the accuracy and efficiency of efficiency measurement and prediction. By integrating NNs, the stochastic DEA model can capture the complex and non-linear relationships between the input and output variables of the decision-making units (DMUs) and handle uncertainty in the input and output data. This is achieved by using the NN to estimate the distribution of the input and output data and then using the stochastic DEA model to calculate the efficiency scores based on these estimated distributions. Furthermore, the integration of NNs into the stochastic DEA model allows for the development of hybrid models that combine the strengths of both techniques. For example, some researchers have proposed using genetic algorithms or other optimization techniques to optimize the input and output weights of the stochastic DEA model, which are then used to calculate the efficiency scores based on the estimated distributions from the NN. This results in a more accurate and efficient efficiency measurement and prediction model. Another novelty of integrating NNs into the stochastic DEA model is the potential for enhancing the interpretability of the model. While NNs are often considered as black-box models, several methods have been proposed to enhance the interpretability of NN-based stochastic DEA models. These methods include using feature importance analysis or visualization techniques to identify the most important input and output variables that contribute to the efficiency scores. Overall, the integration of NNs into the stochastic DEA model represents a novel approach to addressing the limitations of traditional DEA models and improving the accuracy and efficiency of efficiency measurement and prediction under uncertainty. The development of hybrid models and methods to enhance interpretability further add to the novelty and potential impact of this research.