Research On Fuzzy Clustering Algorithm Of Sample And Feature Weighting

Clustering is an unsupervised learning method by which a large amount of knowledge can be obtained from the data.To this end,researchers propose some different clustering algorithms,among which clustering algorithms which based on the objective function are the hotspot of people's research,and they are widely used in some fields,such as pattern recognition,image segmentation,market research,data mining and so on.However,for some proposed clustering algorithms which based on the objective function,they regard the different samples and features as equally important,whicn results in lower clustering performance.In order to further improve the performance of the clustering algorithms,it is of great significance to study the fuzzy clustering methods of different contributions of different samples and features to clustering.Through the analysis and classification research of clustering algorithms,the main contents of this paper are as follows:1.Fuzzy clustering of sample and feature weightingThrough the deep study of the fuzzy clustering of sample weighting and the fuzzy clustering of feature weighting,the fuzzy clustering model of sample and feature weighting based on the objective function of FCM clustering and the normalization of membership degree by taking into account the influence of the importance of samples and features on clustering has been obtained.Then the fuzzy membership degree,the clustering center,the sample weight and the feature weight are theoretically derived by the Lagrange method,and the fuzzy clustering algorithm of dynamic adjusting of sample and feature weights is given.At the same time,we studied the fuzzy clustering algorithm of static adjusting sample and feature weights.In addition,we studied the fuzzy clustering algorithm of the sample and feature weighting which relaxing membership degree normalization constraint to solve the influence of noise data on clustering.2.Kernel fuzzy clustering of sample and feature weightingIn order to solve the clustering of complicated data,combining with the kernel method and considering the two cases of membership degree normalization and relaxing membership degree normalization constraint,we obtained the kernel fuzzy clustering model of sample and feature weighting.The fuzzy membership degree,the clustering center,the sample weight and the feature weight are theoretically derived and the corresponding fuzzy clustering algorithms are given which include the kernel fuzzy clustering algorithms of dynamic and static adjusting sample and feature weights.3.Neural network implementation of fuzzy clustering algorithmIn order to solve the clustering problem of complicated objective function,for the fuzzy C-means clustering,Hopfield neural network and multi-synapses neural network are used to solve the optimization problem of the fuzzy C-means clustering,Hopfield neural network mainly solves the clustering center and multi-synapses neural network solves fuzzy membership degree.On the basis,we proposed neural network algorithm of fuzzy C-means clustering.4.The performance of different methods by the experimental studingThe performance of fuzzy clustering algorithms of sample and feature weighting and their kernel fuzzy clustering algorithms are studied by the experiments which selecting the standard data sets in the UCI database and artificially generated data set,and compared with the commonly used clustering methods by experiments.In addition,the neural network algorithm of fuzzy C-means clustering is studied by using iris data set and artificially generated data set.