Support Vector Machines for Incident Detection in Urban Signalized Arterial Street Networks

An important method to solve the urban traffic congestion is to detect and identify the incident state before it becomes severity. This paper describes the development of support vector machines for urban signalized arterial streets incident detection. Input vector using two types of data: fixed detectors and probe vehicles. Incident detection is accomplished using five approaches: processing traffic input data with ARFIMA model, source data training with SVM, incident state that using to training SVM with fuzzy logic and then multiple attribute of incident state from fixed detector and probe vehicles with data fusion to decide the links and network state. Analysis data generated from a simulation of a small network are used. Different model are used to compared and evaluate the performance of the model of this paper.