Towards Comprehensive Privacy Protection in Data Clustering

Abstract

We address the protection of private information in data clustering. Previous work focuses on protecting the privacy of data being mined. We find that the cluster labels of individual data points can also be sensitive to data owners. Thus, we propose a privacy-preserving data clustering scheme that extracts accurate clustering rules from private data while protecting the privacy of both original data and individual cluster labels. We derive theoretical bounds on the performance of our scheme, and evaluate it experimentally with real-world data.