Kernel K-means Clustering Algorithm Based on Parzen-window Estimation

Aiming at the problem that the initial cluster centers randomly selected in kernel k-means clustering are likely to cause algorithm failure, a method of clustering center determination based on density estimation is proposed. This method uses the idea of kd-tree to choose the point with large probability density and separation as the initial point, and uses the wavelet kernel to describe the nonlinear mapping. Experiments on artificial data sets and real data sets show that the error rate of this method is low and has certain practical significance.