The Spatiotemporal Interplay of Regularity and Randomness in Cellular Data Traffic

(6-page paper); International audience; In this paper, we leverage two large-scale real-worlddatasets to provide the first results on the limits of predictabilityof cellular data traffic demands generated by individual usersover time and space. Using information theory tools, we measurethe maximum predictability that any algorithm has potential toachieve. We first focus on the predictability of mobile trafficconsumption patterns in isolation. Our results show that it istheoretically possible to anticipate the individual demand witha typical accuracy of 85% and reveal that this percentage isconsistent across all user types. Then, we analyze the jointpredictability of the traffic demands and mobility patterns. Wefind that the two dimensions are correlated, which improves thepredictability upper bound to 90% on average.