Detecting Aggregate Bursts from Scaled Bins within the Context of Privacy

In this paper, we consider burst detection within the context of privacy. In our scenario, multiple parties want to detect a burst in aggregated time series data, but none of the parties want to disclose their individual data. We introduce two data perturbation approaches that alter the local data so that raw time series data values are not shared and bursts can be identified using a Shewhart threshold. The first involves lossy data compression via windowing. Unfortunately, windowing alone does not guarantee enough privacy because the envelope of the time series can still be determined. Therefore, we introduce a second data perturbation approach that employs scaled binning. This method transmits values for each data point based on the distance of the data point to a local mean of the time series. The strength of this approach is its increased privacy. We empirically demonstrate the burst detection results using both real and synthetic distributed data sets. When attempting to optimize both privacy guarantees and burst detection accuracy, we find that a combined approach using both windowing and scaled binning balances burst accuracy and privacy better than either approach individually.