<p>Oxygen-18 is an important indicator determining groundwater origin, recharge sources, and age. However, compared to basic geochemical parameters such as major ion composition, oxygen-18 is measured significantly less. This study aims to develop a machine learning (ML) model to successfully predict oxygen-18 values based on ionic composition (Na⁺, K⁺, Mg²⁺, Ca²⁺, Cl⁻, SO₄²⁻, HCO₃⁻), coordinates, well depth as input variables for the whole Baltic Artesian Basin (BAB). A dataset of 567 distinct sample entries was developed from previous research and databases of Lithuania, Latvia, and Estonia. Twelve individual ML models were tested in this research. The prediction results of each model were evaluated using three performance metrics, r-square (R²), mean absolute error (MAE), and root mean square error (RMSE). Overfitting was also evaluated by considering the error metric results of train and test sets and correlation plots of oxygen-18 predicted vs. actual values. The best-performing models—Gradient Boosting, Random Forest, and K-neighbors regressors—achieved R² values greater than 0.8. However, overfitting is observed during the ML of Gradient Boosting and Random Forest models. Hyperparameter tuning helped to increase the accuracy of K-neighbors’ regressor performance without creating overfitting. The study results show that the tuned K-neighbors regressor performance is the best fit: R² 0.82–0.84, MAE 0.98–0.99‰, RMSE 1.67–1.74‰. This study demonstrates that machine learning can be successfully applied to predict oxygen-18 values in groundwater across a basinal scale.</p>