An Improved Selective Ensemble Learning Method for Highway Traffic Flow State Identification

Abstract

Reliable and accurate real-time traffic flow state identification is crucial for an intelligent transportation system (ITS). This identification is a prerequisite for alleviating traffic congestion and improving highway operation efficiency. In this paper, we propose an improved traffic flow state identification model that is based on selective ensemble learning (SEL). First, we adopted the fuzzy C-means (FCM) clustering method to divide the traffic flow data into three main kinds of traffic flow states and obtained the parameters that correspond to each kind of traffic flow state. Second, we applied the random subspace (RS) algorithm as the ensemble method and support vector machine (SVM) model as base learners to construct the RS-SVM ensemble model for traffic flow identification. Significantly, the discrete binary particle swarm optimization (BPSO) algorithm with global optimization search ability was employed to select the classifiers obtained by the random subspace training in the ensemble system. We experimentally validated the effectiveness of the proposed BPSO–RS-SVM-SEL approach. The research results reveal that compared with other classical traffic flow state identification methods, the proposed model has a higher maximum accuracy of 98.68%. It can be seen that our model improves the classification accuracy of traffic flow state identification and the difference in the ensemble system to a certain extent.

Keywords

• Traffic flow state identification
• fuzzy C-means clustering
• random subspace algorithm
• selective ensemble learning
• machine learning