Fluoride pollution is a serious global environmental issue that has attracted widespread attention due to its toxicity, persistence, and tendency to accumulate in ecosystems. In certain lakes and groundwater bodies located in the arid and semi-arid regions of northern and northwestern China, the problem of localized fluoride contamination has become particularly severe, highlighting the urgent need for systematic research and scientific forecasting to identify its causes and understand its evolution. This study focuses on the Daihai Lake Basin in Inner Mongolia, northern China, where 530 samples were collected, including 370 surface water samples and 160 groundwater samples from surrounding areas. By applying hydrogeochemical methods, the study systematically analyzes the hydrochemical characteristics of lake surface water and adjacent groundwater. ArcGIS software was used to map the spatiotemporal distribution patterns of fluoride (F-) concentrations in Daihai Lake and its surrounding groundwater. Correlation and statistical analyses were then performed to investigate the underlying causes and main driving factors of fluoride pollution. On this basis, multiple machine learning models-including K-Nearest Neighbors (KNN), Backpropagation (BP) Neural Network, Gradient Boosting Decision Tree (GBDT), Random Forest (RF), and Grey Relational Analysis-Random Forest-were developed and compared.The results indicate that the surface water in Daihai Lake predominantly exhibits SO4 & centerdot;Cl--Ca & centerdot;Mg and SO4 & centerdot;Cl--Na hydrochemical types, whereas the surrounding groundwater is primarily characterized by the HCO3--Ca & centerdot;Mg type. Surface water fluoride concentrations display clear seasonal variation, following the order: summer > winter > spring > autumn. Spatially, fluoride concentrations are lowest in the western estuary, moderate in the central region, and highest in the northern area. Groundwater fluoride concentrations show no significant temporal variation but exhibit a spatial trend of higher values in the northeast and lower values in the southwest. Fluoride enrichment is mainly controlled by evaporation concentration, fluorite dissolution, and Na+/Ca2+ ion exchange; alkaline conditions with low Ca2+ and high HCO3- concentrations promote fluoride release. The grey correlation-random forest model demonstrated superior performance in predicting fluoride concentrations, achieving RMSE values of 0.916 for surface water and 0.579 for groundwater, and MAE values of 0.690 and 0.429, respectively-significantly outperforming traditional single models. This study provides a scientific basis for the prevention and control of fluoride pollution in the Daihai Basin and offers methodological insights for environmental risk assessment and prediction in similar semi-arid closed-basin lakes.