Short-Term Traffic Flow Forecasting Based on MARS

Abstract

A promising traffic flow forecasting model based on multivariate adaptive regression splines (MARS) is developed in this paper. First, the historical traffic flow data is obtained from the loop detectors installed on the road network of Beijing. Then, part of the data is selected for training the MARS model while the rest is used to test the method. The results based on MARS method are compared with those of other methods such as the neural networks. The proposed MARS method is proved to have a considerable accuracy. Moreover, the model constructed with MARS can be described with analytical functions, which helps a lot in the further research on traffic flow forecasting.