Travel mode classification based on GNSS trajectories and open geospatial data

Abstract

Accurate travel modes inferred from global navigation satellite system (GNSS) trajectory data can be instrumental to city development, such as travel planning and modeling and traffic flow prediction. With the advancement of mobile sensors, the Internet, and GNSS devices, massive GNSS trajectories have been recorded, laying a foundation for travel mode classification at a fine granular scale. However, the lack of discriminative features reduces the accuracy and robustness of travel mode classification. Thus, in this study, considering the relationship between trajectories and the surrounding transportation environment, a set of discriminative features extracted from geospatial data combined with various GNSS features generated from GNSS trajectories is proposed to better distinguish different travel modes. Based on this, we conduct a systematic comparison of a group of state-of-the-art methods using GeoLife and OpenStreetMap (OSM) data, the results of which will provide guidance for properly selecting models for future travel mode classification-related work. In addition, the comparison results show that adding GIS-based domain expert features is robust in improving the classification accuracy of all classifiers in this study.