Research On Fuzzy Clustering Algorithm Of Weighted Feature With Structural ?-entropy

Clustering is an unsupervised machine learning method where in pattern recognition,image segmentation,market research,data mining,computer vision,and other fields has important applications,including clustering method based on classification is a research hot spot,such as k-means clustering algorithm and the fuzzy c-means clustering algorithm,etc.For the traditional clustering algorithm,they mainly takes into consideration that the characteristics of all attributes of the data set are regarded as equally important and it makes some unimportant attributes affect the clustering result of sample.In order to improve the performance of the clustering algorithm,the fuzzy clustering algorithm of weighted feature with structural ?-entropy is studied by introducing the feature weights and its structural ?-entropy.The specific research contents are described as follows:1.The fuzzy clustering algorithm of weighted feature with structural ?-entropy is studied.In view of the different effects of different characteristics on the clustering,the fuzzy clustering model of weighted feature with structural ?-entropy is proposed by introducing the weighted feature and combining the structural ?-entropy of the feature weights.The degree of fuzzy membership,clustering center and feature weights of clustering algorithm are derived by using Lagrange method.On the basis of the iterative method,a fuzzy clustering algorithm of weighted feature with structural ?-entropy is proposed.2.The method of solving feature weights is studied.For solving feature weights,the fuzzy clustering of weighted feature with structural ?-entropy uses dynamic and static two ways.The dynamic way mainly uses iterative update formula of feature weights by using Lagrange to obtain in the process of clustering,the static way mainly uses ReliefF algorithm to calculate feature weights and substitutes it into clustering algorithm and the feature weights in the clustering process remains the same.3.The relationship between feature weights and clusters is studied.In order to obtain a better clustering structure,we analyzes the relationship between the feature weights and clusters of the fuzzy clustering of weighted feature with structural ?-entropy,that is,we derives two conditions,including the feature weights' distribution of different clusters is not necessarily the same and the feature weights' distribution of each cluster is the same from the clustering process.Then,we put forward that considering the feature weights' distribution of different clusters is not necessarily the same can obtain a better clustering structure.4.The influence of loosing membership degree constraint on clustering algorithm is studied.When dealing with the data set that contains noise samples,in order to improve the clustering performance,we loose membership degree constraints for the fuzzy clustering of weighted feature with structural ?-entropy and use Lagrange to derive the result.Then,we put forward the loosing membership degree constraint fuzzy clustering algorithm of weighted feature with structural ?-entropy.5.The kernel fuzzy clustering algorithm of weighted feature with structural ?-entropy is studied.In order to solve more complex structure of data clustering,two different types of kernel based on fuzzy clustering model of weighted feature with structural ?-entropy are proposed according to the different characteristics of the clustering center and the kernel function into the fuzzy clustering model of weighted feature with structural ?-entropy.By solving the model,the degree of fuzzy membership,clustering center and feature weights of clustering algorithm are obtained,and kernel based on fuzzy clustering algorithm of weighted feature with structural ?-entropy is proposed.6.Experiments on the fuzzy clustering algorithm of weighted feature with structural ?-entropy and its kernel algorithms are studied in this paper.In these experiments,we choose UCI standard data sets,synthetic data sets and noise data sets,use dynamic and static two ways to obtain the feature weights,loose the membership constraint and use clustering accuracy and Jaccard coefficient two kinds of evaluation clustering index to carry on the experimental study for the proposed algorithms and compare the experiments on FWFCM(Feature Weighting Fuzzy C-Means)and EWFCM(maximum-Entropy-regularized Weighted Fuzzy C-Means)algorithms.