Using machine learning to predict road weather conditions

Abstract

This thesis reviews two different machine learning models in the prediction of road weather conditions. An empirical dataset covering a five-year period was collected from five road weather stations, and based on correlation analysis, seven variables were chosen to be used as the input of the linear regression model and neural network model, whereas road surface and ground temperatures were used as response variables. The models were trained using data from three of the five stations, and the prediction was done for the other two stations. The results indicate a similarity in predictions between both models, and there was no significant advantage of one model over the other. However, the models gave promising results.