An information theoretic privacy and utility measure for data sanitization mechanisms

Abstract

Data collection agencies publish sensitive data for legitimate purposes, such as research, marketing and etc. Data publishing has attracted much interest in research community due to the important concerns over the protection of individuals privacy. As a result several sanitization mechanisms with different notions of privacy have been proposed. To be able to measure, set and compare the level of privacy protection, there is a need to translate these different mechanisms to a unified system. In this paper, we propose a novel information theoretic framework for representing a formal model of a mechanism as a noisy channel and evaluating its privacy and utility. We show that deterministic publishing property that is used in most of these mechanisms reduces the privacy guarantees and causes information to leak. The great effect of adversary's background knowledge on this metric is concluded. We also show that using this framework we can compute the sanitization mechanism's preserved utility from the point of view of a data user. By using the specifications of a popular sanitization mechanism, k-anonymity, we analytically provide a representation of this mechanism to be used for its evaluation.