A clustering algorithm based on Delaunay Triangulation

Most clustering methods require user-specified parameters or prior knowledge to produce their best results, this demands pre-processing or several trials. Both are extremely expensive and inefficient, because the best-fit parameters are not easy to get. This paper presents a new approach (CBDTM) which is on the basis of Delaunay Triangulation. This approach introduces the median length of k-nearest edges as measure to divide edges for each point. The parameters of CBDTM are not specified by users, and the experiment shows to us that it can find different shape clusters not only in different density data sets, but also in data sets with noise. All operations complete within expected time O(nlogn) , where n is the number of the data sets. The performance comparison experiments show to us, CBDTM more efficient and it has better quality than AUTOCLUST.