Research On Non-spherical Cluster Algorithm

Clustering algorithm is an important unsupervised learning algorithm in the field of artificial intelligence,which is widely used in exploratory pattern analysis,grouping and decision making.It plays an important role in data mining,document retrieval,image segmentation and pattern classification.However,existing clustering algorithms for aspheric clusters have high time complexity,many parameters and poor accuracy.To solve these problems,Chameleon clustering algorithm based on grid local density diffusion is proposed in this paper.Neighborhood diffusion labeling is performed according to global density from high to low,so as to obtain subclusters with high accuracy and realize fast and accurate clustering.On the basis of the former,the method of grid connection is introduced,and the subcluster merging clustering algorithm of space connection is proposed.The algorithm fully considers the distribution of data objects in the clustering space and realizes fast and high precision clustering.In addition,this paper applies the spatially connected subcluster merging clustering algorithm to the user portrait field,which proves the practicability of the algorithm.The main contributions of this paper are as follows:(1)Chameleon clustering algorithm based on grid local density diffusion is proposed:Chameleon clustering algorithm based on local density diffusion was designed to solve the problems of low precision and many parameters in the existing clustering algorithms of aspheric clusters.This algorithm generates subclusters through grid iteration,replacing the cut graph of traditional Chameleon algorithm,and can cluster more quickly and efficiently.The effect of complex aspheric clusters can be improved by means of grid diffusion and cluster merging.The effectiveness of the algorithm is verified by experiments on several datasets with complex aspherical clusters.(2)A spatially connected subcluster merging clustering algorithm is proposed:Aiming at the problems of poor accuracy and slow speed of existing clustering algorithms,this paper further improves Chameleon clustering algorithm based on local density diffusion of grid,and proposes a spatially connected subcluster merging clustering algorithm.Chameleon clustering algorithm based on mesh local density diffusion is faster,more accurate and has fewer parameters.It can fully consider the distribution of data objects in the clustering space,and has a good performance against aspheric clusters and data sets with uneven distribution of density.In the experiment,the proposed algorithm and several new clustering algorithms are tested on artificial data sets containing complex non-spherical clusters and high-dimensional data sets with complex manifolds.The results show that the proposed algorithm outperforms other existing algorithms,especially for the non-spherical data sets and non-equilibrium data sets.(3)The proposed spatially connected subcluster merging clustering algorithm is applied to the user portrait field:In order to prove the application effect of the algorithm,the spatially connected subcluster merging clustering algorithm is applied to the field of user portrait.The clustering algorithm is run on the shopping mall user data set,and the shopping mall users are classified according to the clustering results,so as to accurately describe the user characteristics.