An improved Ambiguity<sup>+</sup> anonymization technique with enhanced data utility

Abstract

The main problem in releasing sensitive data to public networks and domains is how to publish data while protecting privacy and permitting useful analysis. This problem is known as privacy-preserving data publishing. In the well-known (α,β)-privacy model, only two types of risks are considered: presence leakage (membership disclosure) by which adversaries may explicitly identify individuals in (or not in) the published dataset, and association leakage (attribute disclosure) by which they may unambiguously correlate individuals to their sensitive information. Ambiguity is an anonymization technique for this privacy model that publishes attributes of tuples in separate tables. The lossy join of these table produce false tuples which inject uncertainty. In this paper, we improve this anonymization technique (Ambiguity+) that publish the frequency of each distinct value in order to preserve better data utility based on the (α,β)-privacy model. Experimental results demonstrate that our work preserves data utility at a satisfactory level and also information loss is considerably decreased.