Real-Time Stream Statistics via Local Differential Privacy in Mobile Crowdsensing

Mobile crowdsensing has enabled the collection and analysis of stream data. However, the direct processing of gigantic stream data will seriously compromise users’ privacy since those stream data involve numerous sensitive information. To address the challenges of the vulnerabilities of untrusted crowdsensing servers and low data utility, we propose an effective real-time stream statistics mechanism that can not only achieve strong privacy guarantees, but also ensure high data utility. We firstly apply local perturbation on each user’s stream data on the client side, which achieves(\omega )-event(\epsilon )-local differential privacy for each user at each timestamp. Then, we propose a retroactive grouping-based noise smoothing strategy that adaptively exploits the time correlations of stream data and smooths excessive noises, thus improving data utility. Finally, experimental results on real-world datasets show the strong effectiveness of our mechanism in terms of improving data utility.