Modelling the effects of anonymization in O/D matrices obtained from mobile phone data

Abstract

Mobile phone data promise a ubiquitous traffic surveillance without the need of additional hardware. Yet, within most European countries, the data have to be anonymized to disallow the recognition of individuals. To understand the effects of applying the anonymization methodology, we use the results of an agent-based demand model as the ground truth and apply a well-described anonymization method to the trips performed by the simulated population. Besides investigating the segmentation of the area and the time aggregations used by the data supplier T-Systems, we iterate over different cell sizes and aggregation times. The results show how the completeness of the reported data deteriorates with a decrease of cell size and time span used for aggregation. As well, the original segmentation of the area using a variable grid shows to be a good compromise between a fine-grained resolution and the reported number of trips.