Differentially private models of tollgate usage: The Milan tollgate data set

Models of vehicle mobility have applicability in fields such as urban planning and environmental studies. Big Data, specifically, spatio-temporal data, enable novel ways of constructing such models, however privacy remains an issue. Unfortunately, there is no known way to anonymize location data since spatio-temporal data is highly unique to individuals and robust to changes over extended periods of time. Unlike simple anonymization, differentially private summaries of data provide a mathematical guarantee of privacy, while maintaining the utility of the mobility data. Using a dataset from Telecom Italia´s Big Data Challenge, we construct a differentially private model of tollgate traffic in the city of Milan. We then use the model to generate synthetic tollgate data and compare its accuracy on various metrics to that of real data. Our findings show that it is possible to create an accurate, differentially private mobility model from this location data that preserves important characteristics of the original data.