Enhancing Short-Term Traffic Forecasting with Traffic Condition Information

Abstract

One of the key functions of traffic management systems is to monitor traffic conditions and detect the presence of conditions that are abnormal or may not be expected. As archiving of traffic data becomes less costly and more commonplace, generation of short-term forecasts of traffic conditions in real-time conditions is also becoming increasingly possible. Use of condition monitoring information can enhance the performance of short-term traffic forecasting procedures. In this study, one of the most studied approaches to forecasting, the nearest neighbor form of nonparametric regression, is coupled with a condition monitoring method that characterizes the extent to which current traffic conditions deviate from those that may be expected based on historical data. When the normalcy-based approach to traffic condition monitoring was used in the selection of observations from a traffic data archive and in the determination of the nearness of those observations to the current condition, the mean absolute percentage errors for two of the four nearest neighbor forecasting procedures were reduced.