Research On Multi-view Fuzzy Clustering Based On Outliers Analysis

Traditional clustering algorithms,clustering the data of one view or the data of all view information together when processing multi-view data,are mostly for single-view data,and cannot effectively integrate the information contained in multiple views.Based on this,multi-view clustering technology has been proposed and developed rapidly.Like the single-view clustering algorithm,multi-view clustering also has a problem: how to reduce the impact of outliers in the data on clustering.The existence of outlier leads to a decrease in the accuracy of clustering,and even cluster splitting and cluster fusion phenomena,which is undoubtedly a huge challenge for the research of clustering algorithms.To address this problem,the main research of this work is as follows:In order to improve the ability of resisting noise,a multi-view fuzzy clustering algorithm is proposed which,inheriting the advantages of multi-view clustering and fuzzy compactness and separation clustering,can collaborate clustering according to the importance of different views and enhance the robustness.In order to validate the effectiveness of this algorithm,four multi-view data sets are selected to carry out experiments.Experimental results show that this algorithm cannot only achieve high clustering accuracy,but also effectively reduce the impact of noise data on clustering results.In order to reduce the interference of outliers,in the multi-view non-negative matrix factorization,a clustering algorithm combining multi-view non-negative matrix factorization is proposed by introducing L21 norm.The proposed algorithm completes the multi-view non-negative matrix factorization task by alternately iterating the view weight vector,the base matrix and the uniform coefficient matrix of different views.After obtaining the results of multi-view non-negative matrix factorization,the clustering algorithm is used to cluster the uniform coefficient matrix to complete the clustering task.In addition to being able to complete clustering tasks effectively,the algorithm also has the following advantages:(1)identifying the weights of different views;(2)identifying outliers and resisting interference from outliers.In order to verify the effectiveness of the algorithm,an artificial data set and four multi-view data sets were selected for experiments.Experimental results show that compared with the other three algorithms,the algorithm can achieve excellent clustering tasks.Aiming at the shortcomings of the standard non-negative matrix factorization algorithm that is difficult to process multi-view data and cannot retain the geometrical structure of the data space,combined with multi-view non-negative matrix factorization clustering algorithm,a multi-view non-negative matrix factorization algorithm of manifold is proposed by introducing Laplacian matrix.In this algorithm,the multi-view uniform manifold is constructed by the following two steps: in the first step,the Laplacian matrix of each view is obtained;in the second step,the final uniform Laplacian matrix is obtained by linear combination.Experimental results show that compared with the clustering algorithm combined with multi-view non-negative matrix factorization,the non-negative matrix factorization algorithm based on manifold further improves the accuracy of the clustering results.