Mutual Correlation-based Optimal Slicing for Preserving Privacy in Data Publishing

Abstract

Privacy preservation is a substantial concern for the organizations that publish/share personal data for informal analysis. Several anonymization algorithms such as generalization and Bucketization are developed as a solution to this Privacy Preserving Data Publishing (PPDP). Latest research has shown that generalization loses significant amount of information, particularly for high dimensional data. However, Bucketization does not prevent membership disclosure. In this paper, we propose a novel approach that makes use of Information Gain of the attributes with respect to sensitive attributes, which gives the effectiveness of an attribute in classifying the data, which is two-way association among attributes. We show that our approach preserves better data utility and has lesser complexity than earlier techniques. Our proposed technique is theoretically analyzed, and mathematical analysis outstrips past works with sufficient experiments.