Roadcast: an algorithm to forecast this year’s road traffic

Abstract

With the ability to accurately forecast road traffic conditions several hours, days and even months ahead of time, both travelers and network managers can take pro-active measures to minimize congestion, saving time, money and emissions. Typically when a forecast of over an hour is needed, a form of historical average is used. However, this method doesn’t account for contextual factors such as public holidays, major events or weather. This paper presents RoadCast, a machine learning algorithm that uses data on contexts to forecast traffic conditions up to a year ahead of time.
Using 111 loop detectors from Southampton, U.K., RoadCast was found to be more accurate than a historical average predictor in predicting flow and average speed by an average mean squared error of 4.4% and 4.0 % respectively. The improvements came from the algorithm’s ability to ‘learn’ the effect of different contexts on local traffic behavior. RoadCast could be incorporated into a number of Intelligent Transport System (ITS) applications, including context aware route guidance, improved scheduling strategy (planned roadworks, public transport, congestion charging etc.) and faster incident detection with fewer false alerts.

More information

Identifiers

ORCID for Benedict Waterson: orcid.org/0000-0001-9817-7119

Catalogue record

Export record