Filtering and windowing mobile traffic time series for territorial land use classification

Analytics of mobile traffic information may take into account the time-series nature of the data itself. When employing mobile traffic data in a predictive setting to derive useful knowledge to characterize the city environment, the most suitable time series processing methods must be identified. In this paper, we propose an approach to process mobile traffic data using specific time series techniques - smoothing, decomposition, filtering, time-windowing - and to establish the best approach to exploit information extracted from those time series to classify land use, according to sensitivity/specificity metrics. We apply our methodology to a large-scale mobile traffic dataset, we assess its feasibility and we discuss the suitability of different methods for land use classification.