Identification of activity stop locations in GPS trajectories by DBSCAN-TE method combined with support vector machines

GPS sensors on smart phones facilitate collecting and analyzing spatial-temporal information of the phone holders. Before carrying further analysis on these data, splitting continuous GPS trajectories is necessary in data processing. This paper proposes a two-step methodology (a density-based clustering algorithm in the first step and SVMs (support vector machines) in the second step) to deal with GPS data without speed or acceleration features which are usually used as key attributes in the rule-based methodologies. Entropy is used as an updated constraint to remove the erroneously identified stopping points, which leads to 1.5% improvement in the overall accuracy compared with an earlier version of the methodology. The output from the first step also make the SVMs have a better performance.