Efficient Near-Optimal Variable-Size Microaggregation

Microaggregation is a well-known family of statistical disclosure control methods, that can also be used to achieve the k-anonymity privacy model and some of its extensions. Microaggregation can be viewed as a clustering problem where clusters must include at least k elements. In this paper, we present a new microaggregation heuristic based on Lloyd's clustering algorithm that causes much less information loss than the other microaggregation heuristics in the literature. Our empirical work consistently observes this superior performance for all minimum cluster sizes k and data sets tried.