Impact of inaccurate data on Differential Privacy

The growing popularity of location-based systems (LBSs), which enable un-trusted servers to easily collect huge amounts of information regarding users’ locations, has recently raised serious privacy concerns. Differential privacy (DP) is one of the most prominent frameworks used to deal with disclosure prevention in publishing a database. However, the current state-of-the-art DP approaches have a serious drawback: they assume the real-world dataset is accurate, i.e., without errors. Due to the inaccurate positioning services and environmental aspects, location data typically contain errors. The impact of errors on the performance of DP has yet to be explored. To bridge the gap between theory and practice in data privacy, in this paper, we preliminarily analyze the impact of errors on DP and present a rigorous proof of the effect. Furthermore, we highlight the current mechanisms that should be evaluated in applying DP to an inaccurate dataset. The experimental results demonstrate that if errors exist, the released data will have a lower privacy guarantee than expected.