Research On Subspace Fuzzy Clustering Algorithm Driven By Viewpoint

With the development of the Internet,data is now becoming more and more diverse and complex.The FCM algorithm has certain limitations on the data set.Usually,the separation between the spherical classes is obvious,and the compact data set within the class performs better.Obviously,some complicated situations cannot be handled well.To this end,we propose two algorithms in this paper,Viewpoint-driven subspace fuzzy cmeans Algorithm(VSFCM)and nuclear subspace possibility C-means fuzzy clustering algorithm(KSPFCM)algorithm.The main research work of this dissertation is as follows:(1)After in-depth investigation of the research status of fuzzy clustering algorithms,we have analyzed several more classic fuzzy clustering algorithms in detail.Describes their advantages,as well as the shortcomings that still exist.For example,the possibility algorithm introduces a typical matrix to solve the effect of noise points on clustering,but its classification accuracy rate on high-dimensional data sets is low,and its performance is not very ideal.Based on these research and analysis,we set out to solve the problems that still exist in the clustering algorithm,and proposed two fuzzy clustering algorithms.(2)Most of the current fuzzy clustering algorithms are sensitive to cluster initialization and have weak adaptability to high dimensions.For this reason,we propose a viewpoint-driven subspace fuzzy C-means algorithm.First,on the basis of the RLM algorithm,a new cutoff distance is proposed,and thus the cutoff distance-induced cluster initialization method CDCI is established as a new strategy for cluster center initialization and viewpoint selection.Secondly,taking the viewpoint obtained by CDCI as the entry point of knowledge,a new fuzzy clustering strategy driven by knowledge and data is formed.Introduce the subspace clustering mode,the fuzzy feature weight processing mechanism,and give the separation formula of the viewpoint optimization between clusters,and put forward a viewpoint-driven subspace fuzzy clustering algorithm based on these.Finally,through comparative experiments with many advanced clustering algorithms in this field,it is verified that CDCI is more reasonable,and at the same time it performs better in the number of iterations and iteration time.(3)In order to improve the noise resistance of the fuzzy clustering algorithm,we introduced the possibility fuzzy clustering algorithm mechanism,and proposed the kernel subspace possibility C-means fuzzy clustering algorithm.The algorithm is a good combination of the kernel fuzzy clustering algorithm,the possibility fuzzy clustering algorithm,and the subspace fuzzy clustering algorithm.The typicality matrix is introduced,the constraints on the typical matrix are relaxed,and it is updated along with the membership matrix,which reduces the sensitivity of the algorithm to noise points.Introduce the kernel distance,map the data points to the kernel space through the Gaussian function,expand the adaptability of the fuzzy clustering algorithm to the data set,introduce subspace clustering,reduce the granularity of the clustering algorithm to the cluster projection subspace,and greatly enhance the height Adaptability of dimensional data.In Chapter 4,comparative experiments are conducted on 3 UCI data sets.The clustering results show that the KSPFCM algorithm performs best in the five clustering indicators.