Constructing passenger profiles over time by clustering smart card data

The aim of this study is to discover regularities and understand behavior patterns of citizens by clustering temporal smart card data. In this paper, we regard every commuters as a vector of H=168 (7 days multiplied by 24 hours) dimensions, which reflects the total number of trip-chains in corresponding period. After normalization of the transformed input vector, von Mises-Fisher(vMF) clustering model is used to construct passenger profiles. The proposed approach is applied to one-month transit smart cards data collected from the urban transport of Shenzhen, China (Shenzhen Tong).It is clear that every travel trip has its own purpose. Similar groups are aggregated into a hierarchy by proposed model that can be multi-level explored and easy to deduce customer's profiles. And the acquired results shed light on macro structure of passenger behaviors ranging from typical commute conducts to diffuse usage of public transport.