A Linear-Time Multivariate Micro-aggregation for Privacy Protection in Uniform Very Large Data Sets

Optimally micro-aggregating a multivariate data set is known to be NP-hard, thus, heuristic approaches are used to cope with this privacy preserving problem. Unfortunately, algorithms in the literature are computationally costly, and this prevents using them on large data sets. We propose a partitioning algorithm to micro-aggregate uniform very large data sets with cost O(n). We provide the mathematical foundations proving the efficiency of our algorithm and we show that the error associated to microaggregation is bounded and decreases when the number of micro-aggregated records grows. The experimental results confirm the prediction of the mathematical analysis. In addition, we provide a comparison between our proposal and MDAV, a well-known micro-aggregation algorithm with cost O(n2).