Anomaly detection mechanisms to find social events using cellular traffic data

The design of new tools to detect on-the-fly traffic anomaly without scalability problems is a key point to exploit the cellular system for monitoring social activities. To this goal, the paper proposes two methods based on the wavelet analysis of the cumulative cellular traffic. The utilisation of the wavelets permits to easily filter “normal” traffic anomalies such as the periodic trends present in the cellular traffic. The two presented approaches, denoted as Spatial Analysis (SA) and Time Analysis (TA), differ on how they consider the spatial information of the traffic data. We examine the performance of the considered algorithms using cellular traffic data acquired from one the most important Italian Mobile Network Operator in the city of Milan throughout December 2013. The results highlight the weak points of TA and some important features of SA. Both approaches overcome the performance of one reference algorithm present in literature. The strategy used in the SA emerges as the most suitable for exploiting the spatial correlation when we aim at the detection of the traffic anomaly focused on the localisation of social events.