Groundwater depletion and increasing water stress in semi-arid regions demand precise identification of high-potential aquifer zones for sustainable water resource management. Traditional groundwater mapping methods often struggle to capture complex hydrogeological relationships. This study investigates the factors influencing groundwater potential within the Modjo River catchment, located in the Main Ethiopian Rift, by employing ensemble machine learning models, including Random Forest (RF), eXtreme Gradient Boosting (XGBoost), Explainable Boosting Machine (EBM), Gradient Boosting, Categorical Boosting (CatBoost), and Adaptive Boosting (AdaBoost). Hydrogeological, topographical, climatic, and land cover variables served as predictors, while model performance was assessed using precision, recall, F1-score, and ROC-AUC. Results indicate that RF, XGBoost, and CatBoost achieved the highest predictive accuracy, particularly in identifying High and Very High groundwater potential zones, with AUC values reaching up to 0.98. Groundwater potential maps generated by all models display consistent spatial patterns, with central and eastern regions showing high availability. SHAP analysis highlighted geology and lineament density as the most influential factors across models. Although the models concur with broad hydrogeological patterns, disparities in finer spatial details reflect the distinct sensitivities of each algorithm. These findings demonstrate the effectiveness of ensemble machine learning techniques in delineating groundwater potential zones and highlight the importance of geological and structural factors in sustainable groundwater management.Graphical AbstractGraphical Abstract Description: This study investigates groundwater potential in the Modjo River catchment, the Main Ethiopian Rift, employing an integrated machine learning approach. The research systematically acquires and processes diverse geospatial data, including Slope, Elevation, Geomorphology, Rainfall, LULC, Soil, Lineament Density, Drainage Density, and Lithology, which are crucial factors influencing groundwater occurrence. These multi-thematic layers serve as input features for advanced machine learning models, including Random Forest, XGBoost, CatBoost, Gradient Boosting, AdaBoost, and the Explainable Boosting Classifier (EBC). The methodology involves several key steps: (1) comprehensive data collection and pre-processing of relevant hydrogeological and environmental parameters; (2) application of EBM, RF, XGBoost, CatBoost, AdaBoost and Gradient Boosting to model the complex relationships between these parameters and groundwater potential; (3) rigorous evaluation of model performance using various statistical metrics, demonstrating their accuracy and predictive capabilities; and (4) generation of a groundwater potential map, delineating areas with high to low groundwater favorability within the Modjo River catchment. The models were evaluated using precision, recall, F1-score, and AUC metrics. XGBoost and CatBoost achieved the highest AUC scores (up to 0.98), indicating strong performance in delineating High and Very High groundwater potential zones. SHAP-based explainability techniques consistently highlighted geology and lineament density as the most influential factors across all models. The generated maps show consistent spatial trends, with central and eastern parts of the basin exhibiting higher groundwater potential. Area-wise analysis confirms that approximately 44-45% of the study area falls within high-potential zones. These findings provide actionable insights for groundwater exploration, borehole site prioritization, and sustainable water resource management.