Rapid population growth, urban expansion, and industrial development have intensified groundwater dependency, especially in semi-arid regions such as Bundelkhand, India. Effective groundwater potential assessment is crucial for sustainable resource planning and drought resilience. This study evaluates and compares six advanced machine learning (ML) algorithms Decision Tree (DT), Random Forest (RF), Artificial Neural Network (ANN), Ridge Regression, Support Vector Machine (SVM), and Least Absolute Shrinkage and Selection Operator (LASSO)-to delineate groundwater potential zones in the Lalitpur district. A comprehensive set of topographic, hydrological, lithological, and remote sensing-derived variables, including elevation, slope, land use/land cover (LULC), Normalized Difference Vegetation Index (NDVI), Topographic Wetness Index (TWI), and lineament density, were integrated for model development. Seventy percent of the data were used for training and thirty percent for validation. Among all models, LASSO and Ridge Regression demonstrated superior predictive accuracy and stability by effectively addressing multicollinearity and capturing complex environmental interactions. LASSO further highlighted NDVI and plan curvature as the most influential factors in groundwater recharge dynamics. Model performance, assessed through accuracy (0.63 for LASSO, 0.52 for Ridge), Kappa coefficient, sensitivity, specificity and balanced accuracy (0.69 for LASSO, 0.66 for ANN), confirmed the robustness of these regularization-based algorithms. The study underscores the potential of integrating Earth observation data with ML techniques to support groundwater mapping and sustainable water management in data-scarce, semi-arid regions.Graphical AbstractThis work provides a quantitative assessment of groundwater potential using suits of machine learning(ML) models in the semi-arid Bundelkhand region of India, supporting sustainable groundwater management. Multi-source variables including complimentary data sources, remote sensing and field measurements were integrated to delineate potential zones. Six ML algorithms (LASSO, Ridge Regression, Decision Tree, Random Forest, Support Vector Machine, and Artificial Neural Network) were applied and evaluated for predictive analysis and ensuring the best model for GW potential mapping. This work document the effectiveness of combining Earth Observations (EOs) datasets with ML down to road mapping of GW availability. ML algorithms a transferable for groundwater assessment in other arid and semi-arid regions.