Groundwater potential zoning using Logistics Model Trees based novel ensemble machine learning model

Abstract

In this work, the main aim is to map the potential zones of groundwater in Central Highlands (Vietnam) using a novel ensemble machine learning model, namely CG-LMT, which is a combination of two advanced techniques, namely Cascade Generalization (CG) and Logistics Model Trees (LMT). For this, a total of 501 wells data and a set of twelve affecting factors were gathered and selected to generate training and testing datasets used for building and validating the model. Validation of the models was implemented utilizing various quantitative indices, including ROC curve. Results of the present study indicated that the novel ensemble model performed well for groundwater potential mapping and modeling (AUC = 0.742), and its predictive capability is even better than a single LMT model (AUC = 0.727). Thus, the CG-LMT is a promising tool for accurately predicting potential groundwater areas. In addition, the potential map of groundwater generated from the CG-LMT model is a helpful tool for better-studying water resource management in the area.