Phoenix: Privacy Preserving Biclustering on Horizontally Partitioned Data

Emerging business models require organizations to collaborate with each other. This collaboration is usually in the form of distributed clustering to find optimal customer targets for effective marketing. This process is hampered by two problems (1) Inability of traditional clustering algorithm in finding local (subspace) patterns in distributed data and (2) Privacy policies of individual organizations limiting the process of information sharing. In this paper, we propose an efficient privacy preserving biclustering algorithm on horizontally partitioned data, referred to as Phoenix, which solves both of these problems. It assumes a malicious adversary model which is more practical than commonly employed semi-honest adversary model. It is shown to outperform traditional K-means clustering algorithm in identifying local patterns. The distributed secure implementation of the algorithm is evaluated to be very efficient both in computation and communication requirements.