Multi-party k-Means Clustering with Privacy Consideration

The k-means clustering algorithm is a widely used scheme to solve the clustering problem which classifies a given set of n data points in m-dimensional space into k clusters, whose centers are obtained by the centroids of the points in the same cluster. The problem with privacy consideration has been studied, when the data is distributed among different parties and the privacy of the distributed data is to be preserved. In this paper, we apply the concept of parallel computing to solve the privacy-preserving multi-party k-means clustering problem, when the data is vertically partitioned and horizontally partitioned respectively among different parties. We present algorithms for solving the problems for these two data partition models that run in O(nk) time and in O(m(k + log(n=k))) time respectively. The time complexities of the algorithms are much better than others without parallel computing.