Neural Network Based Generalized Entropy Fuzzy Clustering Algorithm

Clustering is an important part of the data miming and improving the clustering accuracyhas an important role to the application. Fuzzy clustering algorithm as a clustering has a goodperformance when deal with the data that can not be clearly classified. In the paper, a fuzzyc-means based on generalized entropy objective function is studied. The problems of theobjective function of the generalized entropy coeiffcient and the fuzzy exponential parity andgeneralization are solved through the neural network method and we proposed a generalizedentropy fuzzy clustering algorithm based on neural network. At the same time, we replace theEuclidean distance to the kernel distance in the objective function and propose a generalizedentropy kernel fuzzy clustering algorithm based on neural network.In addition, the generalized entropy fuzzy c-means algorithm is an improving to thefuzzy c-means algorithm that adds the generalized entropy term to the objective function.Thus, we study the role of the generalized entropy term in the fuzzy clustering and influenceof the various parameters to the clustering results in the objective function.In order to further improve the accuracy of clustering and cluster stability, the clusteringensemble algorithms for generalized entropy fuzzy clustering algorithm based on neuralnetwork and its kernel algorithm are studied in this paper. We construct an ensembleframework ’which include the process of members selection and weighted. The ensembleframework used the membership degree which has a better reflect to the fuzzy clustering asthe clustering members. In the clustering members selection phase, we used three exist fuzzyclustering membership selection methods and we proposed a new selection which include twoselection steps and a weighed step. In the end, we used two ensemble functions to deal withthe selected members and obtain the final result.In the experiments, we using artificial data sets and UCI data sets to test the algorithmwe proposed. The experimental results show that the generalized entropy fuzzy clusteringalgorithm based on neural network and its kernel algorithm we proposed can effectively solvethe fuzzy index and generalized entropy coeiffcient parity and generalization problems. Whilethe two unequal, the algorithms can get better clustering results. Besides, the clusteringensemble algorithms test results also proved that the two selection weighted method weproposed in this paper has a good effect to the fuzzy clustering members. And further proofthe clustering ensemble framework we construct is efficient and can get a higher clusteringresult to the algorithms.