<p>This study presents an advanced spatially-informed methodology for modeling groundwater quality, specifically designed to enhance predictive accuracy in heterogeneous aquifer systems. The proposed framework integrates both the spatial distance between monitored wells and the target estimation point, along with corresponding groundwater quality indicators, as essential input features for machine learning models. A comparative analysis was conducted using five sophisticated machine learning algorithms including Gradient Boosting Regression (GBR), Gaussian Process Regression (GPR), K-Nearest Neighbors (KNN), Multi-Layer Perceptron (MLP) and Random Forest (RF), all implemented in Python. Model training and validation were carried out using groundwater quality datasets from two climatically distinct Iranian provinces: Guilan (a humid northern region) and Qazvin (a semi-arid northwestern region). The study examined five spatial neighborhood configurations, ranging from four to eight adjacent wells, across two seasonal intervals, the first and second halves of the year. Models were trained and tested with a 75:25 data split, optimized using k-fold cross-validation, and evaluated with metrics like RMSE, MAE, R², and Nash-Sutcliffe efficiency. Results demonstrated that RF consistently outperformed other models, yielding the lowest prediction errors and highest correlation values across various spatial and temporal conditions. Although RF presents a higher computational complexity, its predictive robustness makes it an invaluable tool for sustainable groundwater monitoring. Additionally, the findings identified five neighboring wells as the optimal spatial configuration, striking the best balance between prediction accuracy and computational efficiency. This optimized framework offers a reliable and scalable solution for groundwater quality assessment, contributing to more effective water resource management strategies in regions with diverse hydrogeological characteristics.</p>