Machine learning based investigation of influence of weather on transport mobility.

Abstract

Current work involves data-driven analysis of travel mobility. For this purpose, traffic counts of cars and bikes at various traffic routes along With the associated weather have been collected in Oslo, Norway. Amongst the 6 machine learning algorithms compared (linear regression, random forest, decision tree, gradient boost, support vector machine and artificial neural network), the random forest model is seen to perform the best with an accuracy score of around 0.96. The results indicate : a) bike traffic is more sensitive to weather than the car traffic, b) bike traffic counts showing a stronger bimodal distribution with the hour of the day for week days than the car traffic counts, thus suggesting a wider car-usage outside the bimodal peak-times. c) Monthly bike traffic counts is influenced by the weather, while the monthly car counts is influenced by the vacation periods. Oppdragsgiver: TØI