Fuzzy Microaggregation for Microdata Protection

Abstract

In this work we describe a microdata protection method based on the use of fuzzy clustering and, more specifically, using fuzzy c-means. Microaggregation is a well-known masking method for microdata protection used by National Statistical Offices. Given a set of objects described in terms of a set of variables, this method consists on building a partition of the objects and then replace the original evaluation for each variable by the aggregates of each partition. This is, the values in a given cluster are aggregated –fused– and used instead of the original ones. As the problem of finding the best partition for microdata protection is an NP problem, heuristic methods are considered in the literature. Our approach uses fuzzy c-means for building a fuzzy partition, instead of a crisp one.

Novel Iterative Min-Max Clustering to Minimize Information Loss in Statistical Disclosure Control

“…Steps similar to the HM method are then applied to generate clusters based on this ordering. Domingo-Ferrer et al [7] proposed a microaggregation method based on fuzzy c-means algorithm (FCM) [1]. This method repeatedly runs FCM to adjust the two parameters of FCM (one is the number of clusters c and another is the exponent for the partition matrix m) until each cluster contains at least k records.…”

“…Various methods for Microaggregation has been proposed in the literature for protecting microdata [3,4,7,8,11,12,20,22]. The basic idea of microaggregation is to partition a dataset into mutually exclusive groups of at least k records prior to publication, and then publish the centroid over each group instead of individual records.…”

“…For example, Domingo-Ferrer and Mateo-Sanz [3] proposed univariate and multivariate k-Ward algorithms that extend the agglomerative hierarchical clustering method of Ward et al [23]. Domingo-Ferrer and Torra [6,7] proposed a microaggregation method based on the fuzzy c-means algorithm [1], and Laszlo and Mukherjee [13] extended the standard minimum spanning tree partitioning algorithm for microaggregation [26]. All of these microaggregation methods build all clusters gradually but simultaneously.…”

New Multi-dimensional Sorting Based K-Anonymity Microaggregation for Statistical Disclosure Control

“…Various methods for Microaggregation has been proposed in the literature for protecting microdata [3,4,7,8,11,12,20,23]. The basic idea of microaggregation is to partition a dataset into mutually exclusive groups of at least k records prior to publication, and then publish the centroid over each group instead of individual records.…”

“…For example, Domingo-Ferrer and Mateo-Sanz [3] proposed univariate and multivariate k-Ward algorithms that extend the agglomerative hierarchical clustering method of Ward et al [24]. Domingo-Ferrer and Torra [6,7] proposed a microaggregation method based on the fuzzy c-means algorithm [1], and Laszlo and Mukherjee [13] extended the standard minimum spanning tree partitioning algorithm for microaggregation [27]. All of these microaggregation methods build all clusters gradually but simultaneously.…”