Hybrid computational intelligence models for groundwater potential mapping

Abstract

Groundwater is the most important natural resource in many parts of the world that requires advanced new technologies for monitoring and control. This study presents a comparative analysis of three novel hybrid computational intelligence models that consist of a base Decision Stump classifier and three ensemble learning techniques, i.e., Rotation Forest, MultiBoost, and Bagging, for the groundwater potential mapping. Ten influencing factors (i.e., slope, aspect, plan curvature, topographic wetness index, rainfall, river density, lithology, land use, and soil) and 34 groundwater wells from the Vadodara district, Gujarat, India, were used to prepare a geospatial database. Using this database, three hybrid groundwater models, i.e., Rotation Forest based Decision Stump, MultiBoost based Decision Stump, and Bagging based Decision Stump, were developed. Based on a variety of performance metrics, it is revealed that the Rotation Forest based Decision Stump model had the best performance, followed by the MultiBoost based Decision Stump and Bagging based Decision Stump models. However, all the novel hybrid computational models presented here provided improved estimates of groundwater potential compared to those in previous studies and are sufficiently general to be used in many different landscapes around the world.