This study investigates fluoride contamination in groundwater, the resulting health risks, and machine-learning-based prediction of groundwater fluoride contamination in Agra City (India). Fluoride concentrations varied from 0.53 to 5.93 mg/L (mean 1.97 mg/L) and in the majority (58.33%) of cases, levels exceeded acceptable limits as per the Bureau of Indian Standards and World Health Organization. The results reveal that Oral Chronic Daily Intake (CDI) values varied from 0.37 to 14.83 mg/day over 48 locations. The average CDI values were 1.38 mg/day for infants, 1.53 mg/day for children, 3.93 mg/day for teenagers and 4.91 mg/day for adults at baseline. Risk assessment (non-carcinogenic) using Hazard Quotient (HQ) indicated that HQ mean values for infants were found to be highest (mean HQ = 3.06), followed by children (1.70), teenagers (1.31) and adults (1.05). We noted that the proportion of subjects with HQ values > 1.0 reached 91.66% of infants, 45.83% of children, 41.66% of teenagers and 39.58% of adults, indicating a high risk to health especially in young populations. Four machine learning models were compared to predict fluoride concentrations, including Artificial Neural Network (ANN), Random Forest (RF), Support Vector Regression (SVR) and Extreme Gradient Boosting (XGB). The ANN showed the lowest RMSE (0.035), MSE (0.001), MAE (0.031) and highest R & sup2; (0.998), followed by the XGB but reduced performance was also observed using RF compared with the ANN. Results indicated significant differences among models (Friedman test p < 0.05), and post hoc analyses showed that ANN produced greater statistics than other methods. The findings highlight an immediate need for measures to reduce risks and show that artificial neural networks (ANN) effectively predict fluoride concentrations in groundwater. [GRAPHICS] .