Comparison analytic network and analytical hierarchical process approaches with feature selection algorithm to predict groundwater quality

Abstract

Groundwater quality assessment is important for potability and for industrial and agricultural uses of water. Groundwater quality is a measure of contamination by chemicals, biological organisms, sediments, and heat. Reductions of groundwater quality in some areas due to high rates of consumption have prompted the identification of regions in which extraction will be focused. In this study, geographic information systems (GIS)-based analytical network process (ANP) and analytical hierarchical process (AHP) multiple-criteria decision-making techniques using a fuzzy-quantifier algorithm were devised to model groundwater quality in northern Fars Province, Iran. Groundwater quality was assessed by measuring calcium (Ca), chlorine (Cl), magnesium (Mg), thorium (Th), sodium (Na), sulfate (SO4), electrical conductivity (EC), and total dissolved solids (TDS). A membership function based on World Health Organization (WHO) groundwater quality standards was used to create a fuzzy map of each parameter in ArcGIS. Fuzzy maps were generated for each layer using a trapezoidal membership function. The AHP and ANP methods provided weights for each layer to generate groundwater quality maps. Using a feature-selection algorithm, the relative importance of the factors that affect groundwater quality was determined. The results show that, according to the AUC values, ANP generates higher accuracy than fuzzy-AHP (0.954 in comparison to 0.845). The feature-selection algorithm indicates that Ca, Cl, EC, and Na had the greatest impact on groundwater quality conditions. Using ANP and selecting the most important factors can be an economical way, in terms of time and money, to produce highly accurate information that can be used to predict local groundwater quality.