Detecting Activity Patterns from Smart Card Data

During the past decades, the modelling of transport demand by activity based methods has gained considerable attention from the scientific community. Such demand models offer a greater modelling flexibility than traditional models, by modelling transport demand as a phenomenon which emerges from the desire to perform activities at different locations, as opposed to more traditional models where an origin destination demand matrix of trips is distributed over different routes and modes. One of the drawbacks of the activity based paradigm is that data related to activities is more difficult to collect than traffic counts. Modern technologies, such as smart card ticketing systems and smart phones, allow us to collect more detailed accounts of the movements of individual passengers. This gives us the possibility to analyse consecutive journeys and therefore the time a passenger spends in a certain location. This information can be very useful from an activity based modelling perspective. In this paper we take an exploratory approach to derive important activity time intervals from smart card data. We apply a clustering algorithm on the intervals observed at individual stations to detect which time intervals capture enough activities. We then construct a tree-based labelling algorithm that allows us to label the activities and analyse activity chains of individual passengers. We count pairs of consecutive activity labels, visualise the results as a network and calculate which triplets of consecutive activities occur most often. Using this approach, we are able to identify activity patterns that differ from the typical time windows associated with home-work activities.