Transportation mode classification
The increasing amount of digital data in urban research has drawn attention in urban data mining. In urban research (e.g., travel studies in urban areas), researchers who conduct paper- and telephone-based travel surveys often collect biased and inaccurate data about their participants' movements. Although the use of GPS trackers in travel studies improve the accuracy of exact participant trip tracking, the challenge of labelling trip purpose and transportation mode still persists. The automation of such a task would bring benefits to travel studies and other applications that rely on contextual knowledge (e.g., current travel mode of a person). In my M.Sc. thesis work, I focus on transportation mode classification. In particular, I develop a system that improves classification accuracy of ground transportation modes (e.g., bus, car, bike, or walk), when compared with existing systems, by uniquely using GPS-, accelerometer-data and bus stop locations in one system. To elaborate, I design new classification features based on Dwell Time History and a Window History Queue, which uses previously encountered data to increase the classification accuracy of current data. The resulting system remains as a semi real-time classification system, and it achieves a high classification accuracy of 98.5%. Additionally, I explore the performance of classifiers by training with different combinations of GPS-, accelerometer- and GIS data.
Data mining, Urban data mining, Ground transportation mode, Transportation mode classification