Differentially private data sets based on microaggregation and record perturbation

Abstract

We present an approach to generate differentially private data sets that consists in adding noise to a microaggregated version of the original data set. While this idea has already been proposed in the literature to reduce the data sensitivity and hence the noise required to reach differential privacy, the novelty of our approach is that we focus on the microaggregated data set as the target of protection, rather than focusing on the original data set and viewing the microaggregated data set as a mere intermediate step. As a result, we avoid the complexities inherent to the insensitive microaggregation used in previous contributions and we significantly improve the utility of the data. This claim is supported by theoretical and empirical utility comparisons between our approach and existing approaches.