FLIP: A Privacy-Preserving Mechanism for Time Series Data
Authors: Tucker McElroy, Anindya Roy, Gaurab Hore; Published in Journal of Machine Learning Research, Volume 24, Issue 111, Pages 1-29, 2023.
Abstract
Ensuring privacy in released data is a crucial objective for organizations that produce data. Over the years, extensive research has been conducted to develop effective privacy mechanisms. One notable approach is the addition of noise with a guarantee of differential privacy. However, there are concerns about compromising data utility when implementing strict privacy mechanisms. This issue becomes particularly significant in correlated data, such as time series data. The addition of white noise to a stochastic process can significantly alter the correlation structure, which is essential for accurate prediction. In this study, we propose the use of all-pass filtering as a privacy mechanism for regularly sampled time series data. We demonstrate that this approach preserves certain types of utility while providing sufficient privacy guarantees for entity-level time series. Numerical studies are conducted to evaluate the practical performance of the proposed method. Additionally, we apply the method to labor force data and compare its utility properties with other competing privacy mechanisms.
[Abstract]