Real-Time Prediction of Arterial Roadway Travel Times Using Data Collected by Bluetooth Detectors

Abstract

Bluetooth detectors are gaining in popularity as a cost-effective technology for acquiring travel time data. The sensors, which identify and record the unique media access control address of Bluetooth-enabled devices, can measure travel time when a device passes through the detection zone of two consecutive Bluetooth detectors. As with other automatic vehicle identification technologies (e.g., toll tags, automatic license plate recognition systems), there is a time lag because the travel time cannot be acquired until the vehicle has passed the downstream detector location. The increasing desire for accurate and timely traveler information and the desire for proactive traffic control present a need for accurate prediction of near-future travel times along roadway corridors. A significant body of literature has focused on this problem for freeways, but little effort has been directed toward signalized arterials. This paper presents a data-driven model for predicting near-future travel times on signalized arterials in real time by using data acquired from Bluetooth detectors. The model uses the k nearest neighbor pattern recognition technique to identify historical data from which an understanding of the near-future traffic patterns can be extracted. Unlike previous efforts, an objective approach was used to determine the variables to include in the k nearest neighbor feature vector and the optimal model parameters. The calibrated model was evaluated through application to a set of field data obtained from Bluetooth detectors deployed on a signalized arterial. The model provides performance improvements of approximately 20% over a benchmark model.