Publishing differentially private datasets via stable microaggregation

In recent years, differential privacy has emerged as one formal notion of privacy. Data release based on differential privacy can help researchers to perform statistical analysis on sensitive data of individuals. To publish differentially private datasets, there is a need for preserving data utility, along with data privacy. However, the utility of differentially private datasets is often limited, due to the amount of noise being added to the results of queries. In this paper, we address this issue by proposing a microaggregation-based framework that incorporates microaggregation and differential privacy into the data publishing process. We formulate a new notion of stable microaggregation to characterize a desired property of microaggregation and further develop a simple yet effective stable microaggregation algorithm. We experimentally verify the utility reduction of our proposed framework on real-world datasets. The experiments show that the proposed framework outperforms the state-of-the-art methods by providing better with-in cluster homogeneity and also reducing noise added into differentially private datasets significantly.