Understanding surface water–groundwater (SW–GW) interactions helps assess the dynamics of environmental pollutants, particularly nitrate. This study applies machine learning approaches to evaluate SW-GW dynamics in southern Alabama, USA, focusing on river systems and adjacent monitoring wells. Long-term nitrate concentration trends are modeled using Long Short-Term Memory (LSTM) networks, Random Forest (RF), and a hybrid Random Forest–Autoregressive Integrated Moving Average (RF–ARIMA) model. Results show that both LSTM and RF–ARIMA effectively capture the temporal variability of nitrate concentrations in Alabama’s rivers and aquifers, highlighting distinct seasonal and spatial patterns. The RF–ARIMA model outperforms others in predicting surface water nitrate levels, while LSTM provides superior performance for groundwater predictions.