Privacy Preserving Clustering

The freedom and transparency of information flow on the Internet has heightened concerns of privacy. Given a set of data items, clustering algorithms group similar items together. Clustering has many applications, such as customerbehavior analysis, targeted marketing, forensics, and bioinformatics. In this paper, we present the design and analysis of a privacy-preserving k-means clustering algorithm, where only the cluster means at the various steps of the algorithm are revealed to the participating parties. The crucial step in our privacy-preserving k-means is privacy-preserving computation of cluster means. We present two protocols (one based on oblivious polynomial evaluation and the second based on homomorphic encryption) for privacy-preserving computation of cluster means. We have