Assessment of the Efficiency of Decision-Making Units by Combining Artificial Neural Networks and Data Envelopment Analysis

Evaluating the performance and efficiency of similar units within an organization using the DEA (Data Envelopment Analysis) model has been a topic of debate among researchers in recent decades. In this research, for evaluation of the performance and efficiency of Provincial Gas Companies in Iran, first CCR (Charnes, Cooper, and Rhodes) Input-Oriented Multiple Model and AP (Andersen-Petersen) Model were analyzed for ranking efficient units in the format of DEA; however, the weakness of models was determined in terms of separating efficiency of companies. In the DEA model, units with an efficiency score of "۱" are not ranked using classical DEA methods; in other words, DEA does not differentiate between such units. To solve this problem, the AP approach is proposed to classify efficient units. This problem is generalizable due to the lower quantity of units compared to the input and output quantities of the model. In the continuation of this study, to address this problem and analyze the efficiency of companies, attitudes including Performance Calculator Neural Networks were employed, utilizing units clustering and attitude in the format of synthetic models of DEA and ANNs (Artificial Neural Networks), referred to as Neuro-DEA. Analytical results of calculating the efficiency of these models indicated the higher power of calculation and separability of the model for companies in terms of efficiency. The superiority of the neural data envelopment analysis model (Neuro-DEA) lies in its ability to minimize inputs to achieve the desired output level. To measure the efficiency of the companies with the research model, first, a suitable neural network model is simulated, and then, based on the initial data, the network is trained using the desired output (calculated by DEA) until it can learn the reference patterns and calculate the efficiency of the units based on it.Evaluating the performance and efficiency of similar units within an organization using the DEA (Data Envelopment Analysis) model has been a topic of debate among researchers in recent decades. In this research, for evaluation of the performance and efficiency of Provincial Gas Companies in Iran, first CCR (Charnes, Cooper, and Rhodes) Input-Oriented Multiple Model and AP (Andersen-Petersen) Model were analyzed for ranking efficient units in the format of DEA; however, the weakness of models was determined in terms of separating efficiency of companies. In the DEA model, units with an efficiency score of "۱" are not ranked using classical DEA methods; in other words, DEA does not differentiate between such units. To solve this problem, the AP approach is proposed to classify efficient units. This problem is generalizable due to the lower quantity of units compared to the input and output quantities of the model. In the continuation of this study, to address this problem and analyze the efficiency of companies, attitudes including Performance Calculator Neural Networks were employed, utilizing units clustering and attitude in the format of synthetic models of DEA and ANNs (Artificial Neural Networks), referred to as Neuro-DEA. Analytical results of calculating the efficiency of these models indicated the higher power of calculation and separability of the model for companies in terms of efficiency. The superiority of the neural data envelopment analysis model (Neuro-DEA) lies in its ability to minimize inputs to achieve the desired output level. To measure the efficiency of the companies with the research model, first, a suitable neural network model is simulated, and then, based on the initial data, the network is trained using the desired output (calculated by DEA) until it can learn the reference patterns and calculate the efficiency of the units based on it.

کلیدواژه ها:

DEA ، CCR ، Anderson-Peterson ، ANNs ، Neuro-DEA ، Neuron ، Synapse