Privately detecting bursts in streaming, distributed time series data

Surprisingly, privacy preservation in the context of streaming data has received limited attention from computer scientists. In this paper, we consider privacy preservation in the context of independently owned, distributed data streams. Specifically, we want to protect the privacy of each individual participant’s data stream while identifying bursts that exist across participant streams. We define two types of privacy breaches, data breaches and envelope breaches. In order to protect individual data, each participant transforms large subsets of the stream into small vectors that approximate the stream. These small vectors are calculated by summing coefficients of wavelet transforms at different resolutions. The participants share their vectors using bursty, self-eliminating noise. The combined participant vectors can then be used to detect bursts. We find that our approach leads to accurate burst detection results with reduced communication costs. We demonstrate these findings using both real and synthetic data.