Differential-Private Data Publishing Through Component Analysis
Xiaoqian Jiang(a),(*), Zhanglong Ji(b), Shuang Wang(a), Noman Mohammed(b), Samuel Cheng(c), Lucila Ohno-Machado(a)
Transactions on Data Privacy 6:1 (2013) 19 - 34
(a) Division of Biomedical Informatics, UC San Diego, La Jolla, CA 92093.
(b) Department of Computer Science, Concordia University, 1455 De Maisonneuve Blvd. W., QA H3G 1M8.
(c) University of Oklahoma, 4502 E., 41st St #4403, Tulsa, OK 74135-2512.
e-mail:x1jiang @ucsd.edu; ; ; ; ;
A reasonable compromise of privacy and utility exists at an 'appropriate' resolution of the data. We proposed novel mechanisms to achieve privacy preserving data publishing (PPDP) satisfying e-differential privacy with improved utility through component analysis. The mechanisms studied in this article are Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA). The differential PCA-based PPDP serves as a general-purpose data dissemination tool that guarantees better utility (i.e., smaller error) compared to Laplacian and Exponential mechanisms using the same “privacy budget”. Our second mechanism, the differential LDA-based PPDP, favors data dissemination for classification purposes. Both mechanisms were compared with state-of-the-art methods to show performance differences.