Some Researching On Fuzzy C-Means And Fuzzy C-Lines Clustering Algorithms

Fuzzy c-means(FCM)cluster algorithm is a widespread algorithm in fuzzy cluster analysis.FCM is an unsupervised cluster algorithm.It needs to know the number of clusters,c,and the fuzzy exponent,m.The choosing of the two parameters decides the result of clustering.If the number of clusters is improper,the result of clustering can not get the natural structure of the data.If the fuzzy exponent is not appropriate,the algorithm can not obtain felicitous clustering.Any change in the parameter c or parameter m may take a big change in the result of clustering.At present,lots of people invest the two parameters by each other.In this paper the two parameters will be optimized and choose together.This paper is engaged in Genetic algorithm(GA)to optimize the two parameters. In order to optimize the two parameters together,there are some main designs as follows.Firstly,code the two parameters in one binary chromosome,and divide the chromosome into two parts.The first part controls the parameter c;the second part controls the parameter m.Secondly,the fitness function is constructed on the base of some famous validity indices.Thirdly,for the sake of the parameter c is no less than 2, check the parameters before calculating the fitness number of the chromosome.Finally, for the two parts of chromosome don't impact of each other in the process of genetic, the Confined-Two-point Crossover method is adopted.The experiments of MATLAB show that when using the two parameters which get by the proposed method,the FCM algorithm get a better clustering result.Furthermore,this paper also studies the fuzzy c-lines(FCL)clustering algorithm. We find that the direction vector s_i has the possibility of no-single on theory.On the other hand,we define a new distance for the linear clustering center,and construct a new validity index function of FCL algorithm on base of the new distance.The experiments prove the effectiveness of the new validity index.