Big data in data envelopment analysis with undesirable outputs based on simulation and environmental-health matching data of Chinese industrial enterprises

There is a growing enthusiasm to combine the data envelopment analysis (DEA) method with the conception of “big data.” This study proposes a new algorithm to deal with massive data for DEA with undesirable output. First, we modified the current weak disposability model to deal with undesirable output and prove its applicability. Then, we introduced the two fastest approaches (build hull approach and pre-score approach) and found that they cannot be directly employed in DEA with undesirable output because the pre-score approach may not be as powerful. We modified these two approaches to make them capable of dealing with undesirable output and integrated them by proposing a new algorithm. In practice, we used simulated data and data of Chinese industrial enterprises including environmental and health indicators, and all the above data meet the requirements of the model. The results show that our approach can not only be applied to deal with undesirable output in DEA, but it also can perform better than the build hull approach in some situations. Our approach also benefits the pre-score approach if it cannot include total efficiency units. Finally, five datasets were collected or generated to test the capability of our approach.