Performance modeling of Indian business schools: a DEA‐neural network approach

Purpose – The main purpose of the present study is to develop an integrated approach combining data envelopment analysis (DEA) and neural network (NN) for assessment and prediction of performance of Indian B‐schools for effective decision making as error and biasness due to human intervention in decision making is appreciably reduced. Design/methodology/approach – DEA, being a robust mathematical tool, has been employed to evaluate the efficiency of B‐schools. DEA, basically, takes into account the input and output components of a decision‐making unit (DMU) to calculate technical efficiency (TE). TE is treated as an indicator for performance of DMUs and comparison has been made among them. A sensitivity analysis has been carried out to study robustness of the ranking of schools obtained through DEA. Finally, NN is used to predict the efficiency when changes in inputs are caused due to market dynamism so that effective strategies can be evolved by the managers with limited available data. Findings – A total of 49 Indian B‐schools are chosen for benchmarking purpose. The average score of efficiency is 0.625 with a standard deviation of 0.175 when Charnes, Cooper and Rhodes (CCR) model is used. Similarly, when the Banker, Charnes and Cooper (BCC) model is used the average score is 0.888 with a standard deviation of 0.063. The rank order correlation coefficient between the efficiency ranking obtained through CCR and BCC model is 0.736 ( p =0.000) which is significant. The peer group and peer weights for the inefficient B‐schools have been identified. This is useful for benchmarking for the inefficient DMUs. They can identify the parameters in which they lack and take necessary steps for improvement. The peer group for the inefficient B‐schools indicates the efficient B‐schools to which the inefficient B‐schools are closer in its combination of inputs and outputs. The TE obtained through DEA is used as output variable along with input variables considered in DEA as input and output parameters in a generalized regression NN during training phase. It can be observed that root mean square error is 0.009344 and 0.02323 for CCR‐ and BCC‐efficiency prediction, respectively, during training. Similarly, root mean square error is 0.08585 and 0.03279 for CCR‐ and BCC‐efficiency prediction, respectively, during testing. Now, individual schools can generate scenario with the data within their control and test their own performance through NN model. Originality/value – This work proposes integration of DEA and NN to assist the managers to predict the performance of an individual DMU based on input consumed and generate various “what‐if” scenarios. The study provides a simple but comprehensive methodology for improving performance of B‐schools in India.