An Empirical Study of Utility Measures for k-Anonymisation

摘要

k-Anonymisation is a technique for masking microdata in order to prevent individual identification. Besides preserving privacy, data anonymised by such a method must also retain its utility, i.e. it must remain useful to applications. Existing k-anonymisation methods all attempt to optimise data utility, but they do so by using measures that do not take application requirements into account. In this paper, we empirically study several popular utility measures by comparing their performance in a range of application scenarios. Our study shows that these measures may not be a reliable indicator of data utility for applications in practice, and how to use these measures effectively must be considered.