Do machine learning methods used in data mining enhance the potential of decision support systems? A review for the urban water sector

Abstract

With sustainable development as their overarching goal, Urban Water System (UWS) managers need to take into account all social, economic, technical and environmental facets related to their decisions. Decision support systems (DSS) have been used widely for handling such complexity in water treatment, having a high level of popularity as academic exercises, although little validation and few full-scale implementations reported in practice. The objective of this paper is to review the application of artificial intelligence methods (mainly machine learning) to UWS and to investigate the integration of these methods into DSS. The results of the review suggest that artificial neural networks is the most popular method in the water and wastewater sectors followed by clustering. Bayesian networks and swarm intelligence/optimization have shown a spectacular increase in the water sector in the last 10 years, being the latest techniques to be incorporated but overtaking case-based reasoning. Whereas artificial intelligence applications to the water sector focus on modelling, optimization or data mining for knowledge generation, their encapsulation into functional DSS is not fully explored. Few academic applications have made it into decision making practice. We believe that the reason behind this misuse is not related to the methods themselves but rather to the disassociation between the fields of water and computer engineering, the limited practical experience of academics, and the great complexity inherently present