River supercooling is a key part of the freeze-up process that drives the formation of frazil and anchor ice, with potential impacts on water intakes and hydropower operations (e.g., obstruction of intake grates or elevation of tailwater levels). Anticipating when supercooling will occur is therefore critical for mitigating these effects. However, most current methods for modelling river temperatures at freeze-up require the development of a 1D hydraulic model which incorporates thermal processes. This requires detailed river survey data (e.g., cross-sections) which are not often available and can be costly and impractical to obtain. To address this, a machine learning (ML) model was developed to predict supercooling in the Pembina River, Alberta. The model analyses and forecasts time-series data relating to local weather conditions and water temperatures to predict if and when supercooling will occur. Training was conducted using 15-minute records of meteorological and hydrological variables, including air temperature, water temperature, wind speed, riverbed fluxes, and radiative fluxes (shortwave and longwave) from the 2023 freeze-up period. Model performance was evaluated using independent data from the 2024 freeze-up period. Once validated, the model could be adapted to serve as a site-specific, decision-support tool for water intake operators and managers of other infrastructure negatively affected by the formation of river ice each year.