Evaluating Fuzzy Clustering Algorithms for Microdata Protection

摘要

Microaggregation is a well-known technique for data protection. It is usually operationally defined in a two-step process: (ⅰ) a large number of small clusters are built from data and (ⅱ) data are replaced by cluster aggregates. In this work we study the use of fuzzy clustering in the first step. In particular, we consider standard fuzzy c-means and entropy based fuzzy c-means. For both methods, our study includes variable-size and non-variable-size variations. The resulting masking methods are compared using standard scoring methods.