Author details

locations collected in the training dataset, but none of which is labelled as home. This location is documented as sports; however, for this particular user, it is likely that this place is a home rather than a recreation site. While further investigation into the above two typical cases is needed before any definite conclusions are drawn, they nevertheless illustrate that our annotation method based on underlying activity and travel behaviour can effectively predict the activities, which are tailored to each individual. A location may have a single or multiple functions, but people visiting there could have different purposes. The match with geographic information alone is not able to identify this distinction. We shall call the location annotation at

In this study, a cell phone location annotation method has been developed based on spatialtemporal regularities as well as sequential information intrinsic to activity and travel behaviour. The method does not depend on additional sensors and geographic details. The data requirement is simple and its collection cost is low. It is also generic to be transferable to other areas. On top of that, the method is independent of precisely geometric positions of individ-

Experiments on the annotation method using data collected from natural phone communication of users have achieved 76.6% prediction accuracy. With this probability, the activity conducted at a location for a user can be predicted by the spatial-temporal features of the visits disclosed by his/her call records. Furthermore, this study also shows the added value of the integration between machine learning methods and underlying activity and travel behaviour

Nevertheless, despite the spatial-temporal regularities, activity locations still share commonalities in these two dimensions at a certain degree. Activity and travel behaviour is not solely decided by spatial-temporal elements, it is also affected by socio-economic conditions. The first improvement in future research should thus take this general background information into account. In particular, to address the potential causes for misclassifications of home and work/ school locations, the annotation should be combined with the information on the number of home and work/school places of users as well as their work sectors and regimes. A broad picture of users' social networks, obtained from direct surveys and/or social networking sites, would strengthen the prediction of social visit activities. For non-work obligatory and leisure activities, the detailed types in each of these two categories should be handled separately, if a sufficient size of training data for the detailed types is available. The second improvement lies in finding an effective way of annotating locations, which are visited for multiple purposes for a particular user. While this study links the most frequent activity to a location, it dismisses additional activity types, which are performed by the user at different parts but within a same

cell. In the training dataset, 5% of all the locations are visited for multiple purposes.

Today when simple phones are still prevalent constituting nearly 85% of total global handsets in use, this research makes undoubtedly an important contribution to the semantic explanation

the individual level as micro-location-annotation.

110 Smartphones from an Applied Research Perspective

6. Conclusions and future research

uals, thus considerably reducing privacy concerns.

when annotating the location traces.

Feng Liu<sup>1</sup> \*, JianXun Cui2 , Davy Janssens<sup>1</sup> , Geert Wets<sup>1</sup> and Mario Cools<sup>3</sup>

\*Address all correspondence to: feng.liu@uhasselt.be

1 Transportation Research Institute (IMOB), Hasselt University, Diepenbeek, Belgium

2 School of Transportation Science, Engineering, Harbin Institute of Technology, Harbin, China

3 LEMA, University of Liège, Liège, Belgium
