Clustering-oriented privacy-preserving data publishing

Privacy-preserving data publishing has attracted considerable research interests in recent years. One of the problems in such practices is how to trade-off between data utility and privacy protection. This problem heavily deteriorates when the published data are used to do cluster analysis; clustering demands differences between singles for grouping while privacy preserving aims to hide single identifications. In this paper, a mixed mode data obfuscation method AENDO is proposed, which provides a tradeoff strategy from a novel view. The underlying principle is to keep nearest neighborhood structures of data points while data are obfuscated. In particular, for each data point, AENDO differentiates its attributes into neighboring dispersed attributes and neighboring concentrated ones. Furthermore, pertinent statistical data substitution and data swapping strategies are applied to these attributes, respectively. An extensive set of experiments on UCI data sets are provided to assess the effectiveness of our solution, including comparing AENDO with RBT which is one of the best methods on maintaining data usability for clustering. Our results demonstrate that AENDO behaves similarly with RBT on maintaining data utility for clustering, while it outperforms NeNDS by a factor of approximate 10%. Meanwhile, it delivers better anti-inferring effect compared with RBT and NeNDS.