Abstract: Aiming at the shortcomings of clustering by fast search and find of density peaks algorithm(DPC),which calculates massive distance between point objects,has high computational-complexity about clustering process,and needs to select the final cluster centers manually,an improved algorithm that choose clustering centers automatically based on density peak and grid(GADPC) was proposed.Firstly,with the idea of Clique algorithm,all data points are mapped to grid clustering with grid objects rather than point objects,in order to reduce the distance computation and clustering complexity of DPC algorithm.Secondly,the decision accuracy of the number of cluster centers is improved so that it can automatically select cluster centers more precisely.Finally,the relative similarity between grid internal points and adjacent grid points is dealt,so that the edge points and noise points can be solved well.Comparing with machine learning synthetic data sets of UEF and UCI natural data sets,the rand index of those data sets shows that the clustering quality of the improved algorithm is not lower than DPC and K-means algorithm when calculating large data sets,and it improves the dealing efficiency of DPC algorithm.

Key words: Data mining,Clustering analysis,Density peak,Grid,Similarity