基于Canopy的KFCM聚类优化算法

KFCM clustering optimization algorithm based on Canopy algorithm

在以模糊集为理论支持的聚类算法中,KFCM(kernel fuzzy c-means clustering)是一种对核函数进行优化的模糊聚类算法。KFCM算法需要人为指定数据的分类个数,对数据噪声敏感会降低其性能,且类边缘数据点相互影响会导致分类错误。针对这些问题,该文提出一种改进的C-KFCM模糊算法,先用Canopy粗聚类算法给出数据集大致的分类数,接着在聚类部分使用KFCM算法。改进了原KFCM算法的隶属度函数,在噪声点和边缘数据的隶属度中引入其邻域数据的隶属度平均值,使数据中的噪声对算法的影响减小或消失。实验结果表明,改进的C-KFCM算法能自动确定分类数,并且与原KFCM算法相比,C-KFCM将平均准确率提高了0.09%,且聚类效果更稳定。

Among the clustering algorithms supported by the theory of fuzzy set,KFCM algorithm(kernel fuzzy c-mean)is a fuzzy clustering algorithm which can optimize the kernel function.Nevertheless,KFCM must manually specify the classification number of data.The efficiency of this algorithm is influenced by error data point sensitivity and the edge data of different categories that are close to each other may influence each other,leading to classification error.As for the questions in this algorithm,an optimized algorithm C-KFCM is proposed in this paper.Canopy coarse clustering algorithm was used to give the approximate classification number of the data set.Next,KFCM algorithm was used for clustering.At the same time,the membership function of the original KFCM algorithm is improved,and the membership average value of its neighbourhood data is introduced into the membership degree of noise points and edge data.The effect of noise on KFCM is reduced or disappeare.In the end,the algorithm experiment consequence manifests that the C-KFCM can spontaneously confirm the number of data categories.A greater advantage than KFCM lies in the algorithm C-KFCM enhanced the average accuracy rate by 0.9%and the clustering performance was more stable.