1) ProRail monitors ice formation on overhead power lines which can disrupt train services. They have observations of distortions in electric signals from 2011-2017 but need forecasts to plan mitigation. 2) The presentation proposes a data-driven approach combining ProRail observations with weather data to predict probability of ice formation. 3) A model was developed using Gaussian processes on exploratory data analysis of ProRail observations and weather variables to classify days as frost or normal. The model shows good results and captures real world weather patterns. Further work will develop a high resolution forecast and physics-based hazard model.