A neuro-symbolic approach to GPS trajectory classification

This paper proposes approaches to GPS trajectory classification problem in the context of the Rio de Janeiro’s public transit system (with hundreds or more classes). We adopt a weightless neural network architecture combined with both spatial partition and multiclass decision graphs, inspired by a neuro-symbolic sense of adding knowledge from the domain as opposed to the use of a raw machine learning. Experimental results show performance boosts when using some of the proposed strategies.