Improved prediction of groundwater potential zones using a stacking machine learning model

Abstract

Sustainable groundwater management in regions with limited data remains a significant challenge. Identifying potential groundwater zones by combining geological and hydrological information is crucial for the effective use and protection of this resource. In the Southwest Shewa Zone of Ethiopia, groundwater is the main source for household, agricultural, and industrial needs. However, limited hydrogeological data often leads to drilling ineffective wells, emphasizing the need for precise, data-driven mapping methods. This study uses a stacked ensemble machine learning approach to identify groundwater potential zones. The framework combines Adaptive Boosting, Random Forest, Histogram-Based Gradient Boosting, and Extreme Gradient Boosting using a meta-learner to improve prediction accuracy and minimize bias. The area was classified into five groundwater potential levels: very low, low, moderate, high, and very high. Results indicate that 55.8% of the area falls within high to very high groundwater potential zones, while 29.8% corresponds to very low to low potential. Model evaluation using recall, precision, F1-score, and Receiver Operating Characteristic (ROC) demonstrates strong predictive capability and reliable class discrimination across all groundwater potential zones.These findings demonstrate the strong predictive performance of the stacked ensemble learning model and provide a scientific basis for identifying groundwater prospecting zones and guiding well siting in the study area.