Privacy-preserving detection of anomalous phenomena in crowdsourced environmental sensing using fine-grained weighted voting

Abstract

This article addresses the problem of preserving privacy of individuals who participate in collaborative environmental sensing. We observe that in many applications of societal importance, one is interested in constructing a map of the spatial distribution of a given phenomenon (e.g., temperature, CO2 concentration, water polluting agents, etc.) and mobile users can contribute with providing measurements data. However, contributing data may leak sensitive private details, as an adversary could infer the presence of a person in a certain location at a given time. This, in turn, may reveal information about other contexts (e.g., health, lifestyle choices), and may even impact an individual’s physical safety. We introduce a technique for privacy-preserving detection of anomalous phenomena, where the privacy of the individuals participating in collaborative environmental sensing is protected according to the powerful semantic model of differential privacy. We propose a differentially-private index structure to address the specific needs of anomalous phenomenon detection and derive privacy preserving query strategies that judiciously allocate the privacy budget to maintain high data accuracy. In addition, we construct an analytical model to characterize the sensed value inaccuracy introduced by the differentially-private noise injection, derive error bounds, and perform a statistical analysis that allows us to improve accuracy by using custom weights for measurements in each cell of the index structure. Extensive experimental results show that the proposed approach achieves high precision in identifying anomalies, and incurs low computational overhead.