Improving traffic prediction by including rainfall data CONFERENCE

Weather conditions tend to have measurable impact on traffic conditions of the roads. This relationship is commonly studied at the network level without explicit explanation of the link performances. Furthermore, existing studies typically use high resolution traffic data which may not be available across the entire network and especially during the adverse weather conditions. In this study we explore the impact of rainfall intensity on low-resolution speed band data. We also test whether this additional information about rainfall may improve the prediction accuracy of data-driven models for individual roads. To do so, we incorporate the information about the rainfall intensity into support vector machine (SVM) prediction algorithm. As a benchmark we only consider temporal features to predict near future traffic conditions during rainy weather. Numerical results for 616 road segments in Singapore confirm that rainfall impacts traffic conditions in terms of decreasing the driving speed. This reduction increases with the rain intensity. Furthermore, the results show that additional rainfall data enhances the prediction accuracy for certain number of links; while for the others the rainfall information is not that useful.