Identifying Transportation Modes from Raw GPS Data

Abstract

Raw Global Positioning System (GPS) data can provide rich context information for behaviour understanding and transport planning. However, they are not yet fully understood, and fine-grained identification of transportation mode is required. In this paper, we present a robust framework without geographic information, which can effectively and automatically identify transportation modes including car, bus, bike and walk. Firstly, a trajectory segmentation algorithm is designed to divide raw GPS trajectory into single mode segments. Secondly, several modern features are proposed which are more discriminating than traditional features. At last, an additional postprocessing procedure is adopted with considering the wholeness of trajectory. Based on Random Forest classifier, our framework can achieve a promising accuracy by distance of 82.85 % for identifying transportation modes and especially 91.44 % for car mode.