摘要：Microaggregation is known as a successful perturbative mechanism to realize k-anonymity. The method partitions the dataset into groups of at least k members and then aggregates the group members. These aggregated values are published instead of the original ones. In conventional microaggregation methods, it is desired to produce a protected dataset similar to the original one, so close data records are grouped into the same cluster. Accordingly, the aggregation phase of the algorithms are designed to minimize the sum of within-group squared error (SSE), and therefore a simple arithmetic mean in each group is utilized within the aggregation phase to compute the centroids. However, this trivial approach does not consider the proximity of the published values to the original ones, so intruders are able to limit the range of the original values with respect to published data. In this paper, a proximity-aware microaggregation post-processing algorithm is proposed that revisits the aggregation step to remedy this deficiency. Additionally, it is possible to consider different levels of minimum required distances between original record values and their corresponding published ones. Empirical results confirm the superiority of the proposed method in achieving a better trade-off point between disclosure risk and information loss in comparison with similar microaggregation techniques.
关键词：Microaggregation; Statistical Disclosure Control; Data Privacy; Masking.
