Traffic Prediction Using Multivariate Nonparametric Regression

Abstract

The efficient control of traffic on motorways or freeways can produce many benefits, including quicker journey times, fewer pollutant emissions, and reduced driver stress. If it were possible to accurately predict the future state of traffic on a motorway, active measures could be taken to forestall congestion and its attendant negative impacts. This paper presents an intuitive method of producing these forecasts using a pattern matching technique. The technique adopted is novel in that it is a multivariate extension of nonparametric regression that exploits the three-dimensional nature of the traffic state. The application and other facets of the technique are illustrated with actual data from the London orbital motorway. The technique is able to produce forecasts for two of the three traffic state variables with reasonable accuracy and is capable of application on site.