“The global water crisis is exacerbated by significant spatial data gaps in key water stress indicators such as Baseline Water Stress (BWS), posing serious challenges for policy-making and water resource management. This study leverages the XGBoost algorithm one of the most efficient machine learning methods for regression modeling to estimate missing BWS values at the global scale. Key predictor variables include soil moisture, climatic factors (precipitation, temperature, and evapotranspiration), land use/land cover, and elevation, derived from the Aqueduct Water Risk Atlas 4.0 and remote sensing datasets. Through comprehensive data preprocessing and hyperparameter optimization, the model explains approximately 71% of the variance in observed BWS values (R² = 0.711), achieving a Mean Absolute Error (MAE) of 0.727 and a Root Mean Square Error (RMSE) of 1.143 on the standardized 0–5 BWS scale—demonstrating competitive performance compared to prior studies (R² range: 0.60–0.75). Feature importance analysis reveals soil moisture as the dominant hydrological integrator (35.41%), followed by climate variables as the primary driver of the hydrological cycle (24.58%), land use/land cover (18.30%) and population density as key anthropogenic factors (16.04%), and elevation as a topographic modulator (5.67%). The reconstructed global BWS map highlights pronounced spatial heterogeneity in water stress: critical hotspots emerge across the Middle East, North Africa, and South Asia, while higher water security is observed in tropical and temperate regions. This model provides a practical tool for policymakers to identify high-risk areas, develop early-warning systems, and support sustainable water planning. The unexplained variance (29%) underscores the need to integrate socioeconomic data and implement local-scale calibration representing a pivotal step toward addressing the global water crisis.