Using Mobile Phones to Forecast Arterial Traffic through Statistical Learning

Abstract

This article introduces the new component of Mobile Millennium dedicated to arterial traffic. Mobile Millennium is a pilot system for collecting, processing and broadcasting real-time traffic conditions through the use of global position system (GPS) equipped smartphones. Two algorithms that use data from GPS equipped smartphones to estimate arterial traffic conditions are presented, analyzed and compared. The algorithms are based on Logistic Regression and Spatio-Temporal Auto Regressive Moving Average (STARMA), respectively. Each algorithm contains a learning component, which produces estimates of spatio-temporal parameters for describing interactions between the states of arterial links in the network. Additionally, each algorithm contains an inference component, which gives the procedure for processing real-time data into short-term forecasts using these parameters. The algorithms are tested with simulation data obtained from the Paramics software, and from a field test in New York. Both methods provide encouraging results in forecasting arterial traffic conditions using sparse GPS data.