Water stress is a global issue, and it’s getting worse with climate change and the growing demand for water in farming, industry, and cities. Traditional models often fall short—they can’t capture the complex, unpredictable nature of how water stress develops. To tackle this challenge, we need more advanced tools that can predict water stress more accurately and help guide better decisions. This research used five Machine Learning (ML) methods—Support Vector Machine (SVM), Decision Tree (DT), Random Forest (RF), Extreme Gradient Boosting (XGBoost), and integrates a stacked ensemble model as novelty—to see how well they can forecast water stress. Data preprocessing was used, such as median imputation for missing values and feature selection. The overall best performance was achieved by the stacking meta-model that stacked RF, XGBoost, and neural networks with Mean Absolute Error (MAE) = 2.4149, MAE = 1.0388, and R2 = 0.9883. A close second was XGBoost with Root Mean Square Error (RMSE) = 2.4191, MAE =1.0385, and R2 = 0.9883, which was slightly better than RF (RMSE =2.9819, MAE = 1.5158, R2 = 0.9837). These results show that ensemble methods, in particular stacking configurations, significantly enhance predictive power over baseline and individual-approach models, by defining best methods for estimating global water stress.