Application of multiple change point detection methods to large urban telecommunication networks

Abstract

An integral enabler of the smart city vision is the ability to effectively model collective population behaviour. The realisation of sustainable smart mobility is underpinned by the effective modelling of the spatial movements of the population. Furthermore, it is also crucial to identify significant deviations in collective behaviour over time. For example, a change in urban mobility patterns would subsequently impact traffic management systems. This paper focuses on the issue of modelling the collective behaviour of a population by utilizing mobile phone data and investigates the ability to identify significant deviations in behaviour over time. Mobile phone data facilitates the inference of real social networks from their call data records (CDR). We use this data to model collective behaviour and apply change-point detection algorithms, a category of anomaly detection, in order to identify statistically significant changes in collective behaviour over time. The result off the empirical analysis demonstrate that modern change point detection can accurately identify change points with an R2 value of 0.9633.