Multivariate equi-width data swapping for private data publication

Abstract

In many privacy preserving applications, specific variables are required to be disturbed simultaneously in order to guarantee correlations among them. Multivariate Equi-Depth Swapping (MEDS) is a natural solution in such cases, since it provides uniform privacy protection for each data tuple. However, this approach performs ineffectively not only in computational complexity (basically O(n 3) for n data tuples), but in data utility for distance-based data analysis. This paper discusses the utilisation of Multivariate Equi-Width Swapping (MEWS) to enhance the utility preservation for such cases. With extensive theoretical analysis and experimental results, we show that, MEWS can achieve a similar performance in privacy preservation to that of MEDS and has only O(n) computational complexity.