A Collaborative Kernel Clustering Algorithm for Non-Linear Data in Peer-to-Peer Networks

Clustering is a process of dividing data according to their similarity. The existing collaborative clustering methods do not consider the idea of kernel, which are more suitable to deal with segmentation of linear data. In this paper, a distributed k-means algorithm based on the kernel method for nonlinear data is proposed. In addition, considering of the role of attribute weight, the weights are defined for attributes in the presented clustering approach. Experiments on “aspherical” synthetic datasets and real-world datasets show that our method achieves better results, simultaneously, attribute weighting reveals important feature subsets.