Methodology for identifying activities from GPS data streams

Abstract: When the global positioning system became available for civil uses in the early 1990s, there was an enthusiasm and anticipation that information stored in GPS data streams would replace the traditional data collection methods, especially in the transportation field. Despite the wealth of GPS surveys available to practitioners to work with, the existing studies have not made much progress to deliver models for identification of activities from GPS data streams. The lack of models for identifying activities prevents the reconstruction of activity patterns stored in GPS data streams. The present study proposes a methodology for the identification of activities using a rule-based and discrete choice modeling. This novel approach uses a rule-based model that implements the properties of home-based tours in the form of the feedback loop in order to allow identification of home activities. This model is inert to the presence of travel characteristics as it can be applied to most multi-day GPS data sets, and not just prompted recall surveys. In regard to the non-home activities, a discrete choice model is calibrated to Transportation Tomorrow Survey (TTS), for identification of work and other activities. The estimated results are positive, as they are compared against the TTS, and are consistent with the observed patterns.