Investigating route choices and driving behavior using GPS-collected data

Date of Completion

January 2005


Engineering, Civil




The overall objective of this research is to collect real world travel route data using Global Positioning System (GPS) receivers and to develop the models needed to use these data in route choice and other travel behavior research. To achieve the goal three specific analyses are conducted. First, a GIS model was developed to divide the data stream recorded by the in-vehicle GPS receivers into individual trips with the start and end point of the trip being specifically identified. Second, a spatial model was developed to change the typology of the routes (or trips) from representation as a series of points into a series of continuous network links. Automating this data processing will allow analysis of larger datasets for more generalizable results. Third, travel time on each road link in the entire network was estimated using the sparse sample of GPS travel data (256 vehicles each for 10 days spreading over the 18 month study period) as travel time probes. This model is necessary so that the link travel times on each alternative routes faced by the drivers for each trip are known by researchers. This knowledge of the full network travel times, which has not been available in any previous research, will allow for the generation of alternative routes and comparison with the chosen routes to determine the relative influences of different factors on route choice. ^ One specific unique aspect of this work is that data for calibration and evaluation of models were available. The evaluations of the models indicated which combination of parameters was best. The trip dividing model correctly identified 94% of the trips. The accuracy level of the point-to-link data conversion model was 95%. The average difference between the GPS recorded travel time and the estimated travel time for a trip is 70.8 seconds for the 12, 767 trips (average trip length 5,226 meters). Overall, this research provides the first highly reliable and fully evaluated series of GIS models to automatically process GPS collected travel route data. The results will increase the confidence and reliability of GPS usage for route choice research and other transportation planning exercises. ^