Freeway Traffic Speed Estimation by Regression Machine-Learning Techniques Using Probe Vehicle and Sensor Detector Data

Abstract

In the literature, machine-learning techniques have been extensively implemented to capture the stochastic characteristics of freeway traffic speed. The deployment of intelligent transportation systems (ITSs) in recent decades offers much enriched and a wider range of traffic data, which makes it possible to adopt a variety of machine-learning methods to estimate traffic speed. However, an understanding of what type of machine-learning models to select for such applications and how to use probe vehicle data to estimate traffic conditions are still lacking. To fill this research gap, this study aims to utilize regression machine-learning algorithms to estimate traffic speed using probe vehicle and sensor detector data; also, the performance of the utilized machine-learning algorithms is compared using a novel traffic speed estimation framework. The results show that the proposed framework can effectively capture time-varying traffic patterns and has a superior ability to accurately estimate traffic speed in a timely manner. Using sensor detector data as the benchmark, the comparison results show that a random forest achieves the best performance in terms of traffic speed estimation.