Sustainable groundwater management is crucial for safe drinking water in fluoride-affected regions. This study develops a machine learning framework to classify groundwater fluoride using physicochemical predictors. Key features were identified through hydrogeochemical, spatial, and SHAP-based analysis. Model performance was evaluated using nested cross-validation and validated with an independent geology-informed spatial holdout dataset. Multiple machine learning algorithms were tested, including Decision Tree, Support Vector Classifier, Logistic Regression, Gaussian Naïve Bayes, XGBoost, and K-Nearest Neighbours. The Support Vector Classifier outperformed other models, achieving an accuracy of 0.75 and an AUC of 0.821 on the independent spatial test set. The generated prediction maps effectively identify fluoride-prone zones, providing a practical tool for groundwater management, risk mitigation, and sustainable planning in water-stressed regions. This study also supports Sustainable Development Goal 6 by contributing to improved water quality and sanitation strategies.