Efficient distributed privacy preserving clustering

Official URL: http://192.168.1.20/record=b1226354 (Table of Contents)

With recent growing concerns about data privacy, researchers have focused their attention to developing new algorithms to perform privacy preserving data mining. However, methods proposed until now are either very inefficient to deal with large datasets, or compromise privacy with accuracy of data mining results. Secure multiparty computation helps researchers develop privacy preserving data mining algorithms without having to compromise quality of data mining results with data privacy. Also it provides formal guarantees about privacy. On the other hand, algorithms based on secure multiparty computation often rely on computationally expensive cryptographic operations, thus making them infeasible to use in real world scenarios. In this thesis, we study the problem of privacy preserving distributed clustering and propose an efficient and secure algorithm for this problem based on secret sharing and compare it to the state of the art. Experiments show that our algorithm has a lower communication overhead and a much lower computation overhead than the state of the art.