Modeling the Various Merging Behaviors at Expressway On-Ramp Bottlenecks Using Support Vector Machine Models

Abstract

Abstract Merging behaviors on the acceleration lane are viewed as a key trigger in expressway breakdown and potentially increase driving risk. The main goal of this article is to make a high-precision prediction on four kinds of merging behaviors at expressway on-ramp bottlenecks. Based on the videos of traffic flow at two on-ramp bottlenecks at Yan’an Expressway in Shanghai, 403 empirical samples are collected by extracting trajectories from merging vehicles, as well as each adjacent one using trajectory processing software. A learning-based support vector machine (SVM) approach is adopted to predict the various merging behaviors. Considering four merging behaviors have different degrees of merging risk and effect on the mainline traffic flow, three SVM models are established. To overcome the potential over-fitting problem, variance analysis is used to extract the key variables in each model. The results show that the SVM models perform well. The highest prediction accuracy in the binary-classification models reaches up to nearly 90% followed by 80.10% in the multi-classification model. Besides, the results of SVM models are compared with several frequently used models, including discrete choice model, Bayesian network and classification and regression tree. It turns out that SVM achieves the best prediction results. The proposed method can be used for traffic simulation and real-time driver assistant system on future automated and connected vehicles.