Estimating and predicting journey times from historical HATRIS data

The Highways Agency Traffic Information System (HATRIS) contains average journey time for every junction-to-junction link on the HA network, for journeys starting every 15 minutes since September 2002. There is a need to estimate missing records and predict the values of future observations. This paper details the process of reviewing the current estimation and prediction practices and the development and testing of improved methods. Analysis, including a k-means clustering algorithm, is used to identify natural groupings in the data leading to the recommendation of a revised method of selecting historical data for use in estimation and prediction. A quantitative assessment of the accuracy of a number of methods, simulated using a software tool, identifies an improved averaging method with which to combine these data and arrive at the best estimate of a given journey time.