Objective <p>Nitrate contamination of groundwater is hazardous to humans and ecosystems, necessitating initiatives focused on predictive monitoring frameworks. In this study, we attempted to integrate machine learning and explainable artificial intelligence to predict nitrate levels and to highlight the spatial effects on hydrogeochemical interactions.</p> Data/Methods <p>Using Central Groundwater Board datasets (2019, 2023), we compared 19 ML regression models with/without spatial coordinates (longitude, latitude). Top models (LightGBM, Linear Regression) were interpreted using SHAP, with a novel SPAMAXAC index that combines spatial weights, deflated SHAP R² (DRRXA), feature interactions (FIC), and inflation diagnostics (IAI).</p> Main findings <p>ML achieved moderate R² (0.44–0.51), with chloride/bicarbonate dominating under spatial effects, shifting to phosphate/pH (2019) or chloride/sodium (2023) without spatial effects. SHAP inflated R² by 25–28%, but SPAMAXAC scores (0.52–1.23) revealed strong spatial autocorrelation in 2019 and minimal spatial autocorrelation in 2023, thereby correcting interpretability biases.</p> Implications <p>SPAMAXAC enhances ML-XAI reliability for groundwater studies by penalizing spatial artifacts and SHAP inflation, offering a diagnostic framework beyond standard prediction</p>