Assessment of machine learning models in GIS for predicting groundwater in semi-arid regions of eastern Iran

Predicting groundwater potential is crucial for systematic development and planning of water resources. The main objective of this study is to develop machine learning models including Random Forest (RF), Decision Tree (DT), and Support Vector Machine (SVM) for predicting potential groundwater areas in the Birjand plain. Therefore, for the implementation of this study, geohydrological data related to 37 groundwater wells (including the number and location of wells and groundwater levels) and 17 hydrological, topographical, geological, and environmental criteria were used. Feature selection was performed using Support Vector Machine's least squares method to determine effective criteria for improving the performance of machine learning algorithms. Ultimately, predictive maps of groundwater potential were prepared using DT, RF, and SVM models, and the performance of these models was evaluated using the Area under the Curve (AUC) and other statistical indicators. The results showed that the DT model (AUC=0.89) has very high predictive capability for groundwater potential in the study area, and elevation was identified as the most important factor in predicting groundwater potential in this area. The findings of this study can serve as a guide for decision-making and appropriate planning in the optimal use of groundwater resources.