Travel mode classification of intercity trips using cellular network data

Abstract Many applications in transport planning require an understanding of travel patterns separated by travel mode. To use cellular network data as observations of human mobility in these applications, classification by travel mode is needed. Existing classification methods for GPS-trajectories are often inefficient for cellular network data, which has lower resolution in space and time than GPS data. In this study, we compare three geometry-based mode classification methods and three supervised methods to classify trips extracted from cellular network data in intercity origin-destination pairs as either road or train. To understand the difficulty of the problem, we use a labeled dataset of 255 trips in two OD-pairs to train the supervised classification methods and to evaluate the classification performance. For an OD-pair where the road and train routes are not separated by more than four kilometers, the geometry-based methods classify 4.5% - 7.1% of the trips wrong, while two of the supervised methods can classify all trips correctly. Using a large-scale dataset of 29037 trips, we find that separation between classes is less evident than in the labeled dataset and show that the choice of classification methods impacts the aggregated modal split estimate.