| K-means clustering algorithm is a kind of partitioning algorithm,so there are some shortcomings.In order to solve this problem,the fuzzy C-means algorithm introduces the membership degree and solves these problems,but the fuzzy C-means can not initialize the cluster center and number of classes.The affinity propagation algorithm is a new clustering algorithm,which has advantages over traditional algorithms in terms of efficiency.However,when it comes to non-spherical data sets,it can not produce good clustering results.Based on the above problems,the following research aspects are proposed:In order to solve defects that fuzzy Cmeans can not initialize the cluster center and the number of clusters,this paper uses Init-cluster to select the initial cluster center and the number of clusters,When these parameters are determined,the kernel fuzzy C-means algorithm is used for clustering.In order to avoid the influence of outliers on the clustering results,this paper proposes an angle-based anomaly detection algorithm,which can effectively detect outliers.The algorithm can effectively eliminate outliers and improve the accuracy of the algorithm.Faced with the linear indivisibility of high-dimensional data sets,this paper introduces a combination of global kernel functions and local kernel functions.Since different types of kernel functions have different effects,the new kernel functions have two types of kernel function advantages.The affinity propagation algorithm is based on the Euclidean distance to construct the similarity matrix,so the clustering effect on the manifold dataset is not ideal.This paper proposes a similarity measure based on the graph,redefines the similarity,and enlarges the nuances between the elements.We can handle not only ordinary data,but also non-spherical data.In the face of increasing data,how to perform dynamic incremental clustering based on existing clustering results becomes very important.Existing traditional clustering algorithms can only cluster static data points,this article starting from the angle between the elements,another way is to propose an angle-based incremental clustering algorithm(ICABOA)to avoid a lot of double counting.Based on the kernel fuzzy C-means clustering algorithm,this paper proposes the degree of dissimilarity between classes and directly assigns data to corresponding categories according to the degree of dissimilarity,which solves the defect that traditional clustering algorithms cannot cluster dynamic datasets..Experiments show that this saves time and improves efficiency. |
