This paper proposes an Artificial Intelligence (AI)-based method for predicting global water scarcity, which would help make better sustainable water management decisions. Gradient Boosting, Random Forest (RF), Support Vector Machine (SVM), Logistic Regression, and K-Nearest Neighborss (KNNs) are some of the supervised machine learn ing techniques used in the model. To address the natural class imbalance of the dataset itself, as novelty, the method incorporates Synthetic Minority Over-Sampling Technique (SMOTE) to generate synthetic samples, and weighted classification to enhance model fairness. Among the models evaluated, ensemble methods were found to perform better, the most accurate (97.33%) and highest scoring (96.92%) being Gradient Boosting in the F1 score, followed very closely by Random Forest at 96.00% accuracy and 94.37% F1 score. These results verify the efficiency of ensemble methods to handle structured, complex data of skewed distribution. On the other hand, SVM and KNN had poorer performance, where KNN had the worst results (accuracy: 80.67%, F1 score: 78.58%), revealing the incompetence of KNN in learning patterns from the minority classes despite the use of SMOTE. Logistic Regression was then a strong candidate, having achieved an accuracy of 93.33% and an F1 score of 92.45%. This study emphasizes municipal (domestic) and agricultural water use which are the most significant effect on water scarcity classification, followed by larger spatial and temporal measures. This work demonstrates the potential for a complete data pipeline—ranging from preprocessing to validation and deployment of models—of delivering actionable insights into global water stress.