Research On Cluster Center Optimization Of K-means Algorithm

K-means algorithm is a major breakthrough in the field of clustering,and has been widely used in fraud detection,image processing and market analysis due to its convenience.However,in many application scenarios,k-means algorithm will be affected by the random selection of the initial cluster center in the clustering process,and its application effect is not well.Aiming at the above problem of random selection of class cluster center of K-means algorithm,MDP-KM algorithm and KNNDP-KM algorithm are proposed in this paper by improving density peak value.The main contents of this paper are as follows:1.Aiming at the problem of unstable truncation distance in density peak algorithm,MDP-KM algorithm is proposed.In this algorithm,partial samples are randomly selected and their truncation distance is estimated by dichotomy search,and the minimum distance between the local density of the sample and the corresponding sample object is given.The initial cluster center of k-means algorithm is determined by decision graph.2.Aiming at the problem of selecting the initial cluster center of k-means algorithm,KNNDP-KM algorithm is proposed by combining K-nearest neighbor algorithm and density peak algorithm.The algorithm uses K-nearest neighbors to re-estimate the local density in the density peak algorithm,and KNNDP algorithm is used to get the initial class cluster center of K-means algorithm.Simulation and empirical analysis show that MDP-KM and KNNDP-KM algorithm can better determine the initial cluster center of K-means algorithm,and under Silhouette,Dunn and AMI clustering evaluation indexes,MDP-KM algorithm and KNNDPKM algorithm are better than K-means algorithm and WSKM algorithm.Therefore,the two methods proposed in this paper to determine the initial cluster center of k-means algorithm effectively improve the clustering results of K-means algorithm.