Statistical Analysis on Multivariate Expressway Time Series Traffic Under the Different Toll Policies

Abstract

Recently, various toll policies and social experiments have been conducted for ETC users of Japanese expressways. While the vast number of reports refers to the change of traffic for each section, the influence of toll policies has rarely been evaluated from the statistical viewpoint, due to the lack of an adequate model to deal with the multivariate time series including many peaks. The purpose of this study is to clarify the influence of toll policies on the traffic levels on expressways by using Independent Component Analysis (ICA). The continuously observed traffic data at several different sites on the expressway were decomposed into several independent series by using mixing coefficients (i.e. weight parameters) to convert the independent series into the observed series. ICA isolates the common and stable fluctuations in several traffic series in relation to toll policies. Moreover, the estimated independent series under the different toll policies were statistically compared to clarify whether the significant difference between the independent series occurs, or not.