An Emerging Strategy for Privacy Preserving Databases: Differential Privacy

Abstract

Data De-identification and Differential Privacy are two possible approaches for providing data security and user privacy. Data de-identification is the process where the personal identifiable information of individuals is extracted to create anonymized databases. Data de-identification has been used for quite some time in industry to sanitize data before it is outsourced for data-mining purposes. Differential privacy attempts to protect sensitive data by adding an appropriate level of noise to the output of a query or to the primary database so that the presence or the absence of a single piece of information will not significantly alter the query output. Recent work in the literature has highlighted the risk of re-identification of information in a de-identified data set. In this paper, we provide a comprehensive comparison of these two privacy-preserving strategies. Our results show that the differentially private trained models produce highly accurate data, while preserving data privacy, making them a reliable alternative to the data de-identification models.