Research And Improvement Of Peak Density Clustering Algorithm

Clustering is the process of dividing input data into groups in an unsupervised manner.It can identify the clustering structure of the data set.Therefore,it has been applied in biological science,mobile communication,data retrieval,genetics,machine learning and so on.The clustering technique has the characteristics of recognizing arbitrary shape clusters,but it needs manual input parameters and manual selection of clustering centers.The research in this thesis will be able to select cluster centers automatically and reduce the effect of parameters on clustering performance,The main work is organized as follows:Firstly,this algorithm introduces Laplace theory and Laplace energy of graphs.The Laplacian centrality and distance of each data are calculated,the cluster centers are selected according to the principle of normal distribution.Then data objects can be filtered out.The algorithm solves the problem of parameters and the selection of clustering centers manually.It is verified that the algorithm improves the clustering efficiency,In particular,Automatic clustering is realized.Secondly,a algorithm based on PageRank was proposed,the influence index of data objects are defined by the idea of calculating the PageRank score of web pages in PageRank algorithm,and the relative distance of each data object is calculated.Then,the influence index and relative distance of each data object are mapped to the two-dimensional space,the clustering is carried out after selecting cluster center.By comparing the data size and data dimension with the three classical density clustering algorithms,It is verified that the algorithm not only reduces the sensitivity of parameters,but also optimizing clustering performance.