The goal of the ATMo2 research project is to study the effectiveness of low-cost incentives on the willingness of travelers to shift from motorized individual transport to more environment-friendly transport options. Towards this end, the project develops a mobile app that tracks the daily journeys of its users and proposes sustainable alternatives. To realize this app function in a privacy-preserving manner, the app must be able to perform both, tracking and transport mode identification, on the mobile device of each user.
In our upcoming presentation at Wissenschaftsforum Mobilität and the associated publication, we describe and evaluate machine learning models that can be used for such an on-device transport mode classification. As basis for the construction and evaluation of the models, we leverage a dataset containing GPS traces of several thousand trips. The evaluation indicates that, although some modes are harder to differentiate, it is possible to create a robust on-device classifier that suits the needs of ATMo2.