Abstract: The proliferation of mobile smart devices has led to a rapid increase of
location-based services, many of which are amassing large datasets of user trajectory
information. Unfortunately, current trajectory information is not yet sufficiently rich to
support classification of user transportation modes. In this paper, we propose a method that
employs both the Global Positioning System and accelerometer data from smart devices to
classify user outdoor transportation modes. The classified modes include walking,
bicycling, and motorized transport, in addition to the motionless (stationary) state, for
which we provide new depth analysis. In our classification, stationary mode has two
sub-modes: stay (remaining in the same place for a prolonged time period; e.g., in a parked
vehicle) and wait (remaining at a location for a short period; e.g., waiting at a red traffic
light). These two sub-modes present different semantics for data mining applications. We
use support vector machines with parameters that are optimized for pattern recognition. In
addition, we employ ant colony optimization to reduce the dimension of features and
analyze their relative importance. The resulting classification system achieves an accuracy
rate of 96.31% when applied to a dataset obtained from 18 mobile users.