Periodic patterns in human mobility

The recent rise of services and networks that rely on human mobility has prompted the need for tools that detect our patterns of visits to locations and encounters with other individuals. The widespread popularity of locationand encounter-aware mobile phones has given us a wealth of empirical mobility data and enabled many novel applications that benefit from automated detection of an individual’s mobility patterns. This thesis explores the presence and character of periodic patterns in the visits and encounters of human individuals. Novel tools for extracting and analysing periodic mobility patterns are proposed and evaluated on real-world data. We investigate these patterns in a range of datasets, including visits to public transport stations on a metropolitan scale, university campus WLAN access point transitions, online locationsharing service checkins, and Bluetooth encounters among university students. The methods developed in this thesis are designed for decentralised implementation to enable their real-world deployment. Analysing an individual’s visit and encounter events is a challenging problem since the data are often highly sparse. In order to study visit patterns we propose a novel inter-event interval (IEI) analysis approach, which is inspired by neural coding techniques. The resulting measure, IEI-irregularity, quantifies the weekly periodic patterns of an individual’s visits to a location. To detect encounter patterns we propose and compare methods based on IEI analysis and periodic subgraph mining. In particular, we introduce the novel concept of a periodic encounter community; that is, a collection of individuals that share the same periodic encounter pattern. The decentralised algorithms we develop for periodic encounter community detection are of particular relevance to human-based opportunistic communication networks. We explore these communities in terms of their opportunistic content sharing performance. Our findings show that periodic patterns are a prominent feature of human mobility and that these patterns are algorithmically detectable.