Evaluating the efficiency of water development–utilisation–treatment system: a DEA-BPNN model

With rapid economic growth and urbanisation, water shortage and water pollution are becoming more and more serious. It is very important for decision makers to understand the efficiency of the water system and know its development trend. Data envelopment analysis (DEA) is a robust tool for assessing efficiency. However, the DEA model lacks predictive capabilities, and cannot give guidance on future development. In contrast, a back-propagation neural network (BPNN) offers powerful non-linear mapping and adaptive prediction capabilities. To compensate for the deficiencies of the DEA model, a three-stage DEA-BPNN model is developed based on environmental compatibility and economic development. This model enables specific efficiency measurements, identifies system weaknesses and anticipates future trends. Then, the proposed model is applied to a ‘one belt and one road’ region, comparing its predictive performance with that of linear regression, a generalised additive model, support vector machines, k-nearest neighbours, random forest and gradient boost decision trees. Results reveal that following determination by several prediction models, the BPNN model obtains the more accurate prediction results.