A parsimonious model for classifying the traffic state of urban road networks: A two-stage regression approach

Abstract

An effective method of traffic state classification is crucial for managing urban traffic congestion. Existing methods usually assume a given number of state categories, which is not flexible if real applications are required to define different state levels. In this study, a parsimonious statistical model is derived and validated for classifying urban traffic states. The model is developed on the basis of a large-scale empirical travel speed dataset from five cities in China. First, a hybrid clustering method that integrates DBSCAN and natural breaks is used to derive traffic state classification under various numbers of state categories. The classification results are then compiled to conduct the subsequent regression analysis. Second, a two-stage regression approach is proposed to investigate the correlation between the number of state categories and the classification criteria (i.e., state thresholds that separate one state level from another). In the first stage, a significant linear relationship between the classification criteria of adjacent traffic states is derived ($\overline{R^2}$ ​= ​0.80, P ​< ​0.001). In the second stage, a significant correlation between the slope, intercept, and number of state categories is derived ($\overline{R^2}$ ​= ​0.95, P ​< ​0.001). On the basis of the two-stage regression analysis, a novel parsimonious statistical model is developed. Third, the developed model is evaluated with three performance indicators, namely, the mean squared error (MSE), mean absolute error (MAE), and mean relative error (MRE). The claffication accuracy is further validated via a case study on the speed data of Foshan Avenue North road. We suggest that the model can be used to assist flexible decision-making support with different levels of detail.