Clustering Heuristics for Efficient t-closeness Anonymisation

Abstract

Anonymisation based on t-closeness is a privacy-preserving method of publishing micro-data that is safe from skewness, and similarity attacks. The t-closeness privacy requirement for publishing micro-data requires that the distance between the distribution of a sensitive attribute in an equivalence class, and the distribution of sensitive attributes in the whole micro-data set, be no greater than a threshold value of t. An equivalence class is a set records that are similar with respect to certain identifying attributes (quasi-identifiers), and a micro-data set is said to be t-close when all such equivalence classes satisfy t-closeness. However, the t-closeness anonymisation problem is NP-Hard. As a performance efficient alternative, we propose a t-clustering algorithm with an average time complexity of $O(m^{2} \log n)$ where n and m are the number of tuples and attributes, respectively. We address privacy disclosures by using heuristics based on noise additions to distort the anonymised datasets, while minimising information loss. Our experiments indicate that our proposed algorithm is time efficient and practically scalable.