k-Proximal plane clustering

Instead of clustering data points to cluster center points in k-means, k-plane clustering (kPC) clusters data points to the center planes. However, kPC only concerns on within-cluster data points. In this paper, we propose a novel plane-based clustering, called k-proximal plane clustering (kPPC). In kPPC, each center plane is not only close to the objective data points but also far away from the others by solving several eigenvalue problems. The objective function of our kPPC comprises the information from between- and within-clusters data points. In addition, our kPPC is extended to nonlinear case by kernel trick. A determinative strategy using a Laplace graph to initialize data points is established in our kPPC. The experiments conducted on several artificial and benchmark datasets show that the performance of our kPPC is much better than both kPC and k-means.