Estimation of cost-activity function in activity-based costing using combination of neural networks-Multilayer data envelope analysis in Maskan Bank

Activity-based costing since it's introduction has attracted so much attention. There are, however, practical problems in implementing this costing system, which, in spite of the computational superiority of activity-based costing compared to traditional costing, organizations and companies are still not interested in using this costing method. In the present study, implementation problems that are practically related to implementation of activity-based costing have been investigated and artificial neural networks have been used to solve the problem of estimating the cost-activity relationship (CER) as well as reducing the costs of doing timing in organizations. The statistical population of the research is all branches of Maskan Bank which has been clustered using CI-DEA Data Envelopment Analysis (CI-DEA) and based on performance similarity in 1395. 450 branches were selected as samples and used to train and test the model of neural networks. The distinctive feature of this pattern is to consider non-linear relationship between cost-activity and other patterns. The proposed architecture of network makes it possible, in addition to the cost-of-activity forecast, to be extrapolated from the model, the amount of cost-driven input (time) used as a cost-sharing actuator to the activity in the conventional executive model. The results of the RMSE and MAE models showed that the proposed model has the capability to estimate the cost-activity relationship.