基于非欧式距离的可能性C-均值聚类

改进型可能性C-均值聚类(Improved possib ilistic C-means,IPCM)是在综合了模糊C-均值聚类(Fuzzy C-means,FCM)和可能性C-均值聚类(Possib ilistic C-means,PCM)的基础上得到的。在IPCM的基础上,利用鲁棒统计观点和影响函数,引入一种新的距离度量以代替IPCM的目标函数中的欧式距离度量,提出了一种新的可能性C-均值聚类模型(A lternative improved possib ilistic C-means,A IPCM),并给出了该模型的具体实现算法。A IPCM具有良好的鲁棒性,更适合对含有噪声或野值的数据进行划分聚类。仿真实验表明,A IPCM能克服噪声敏感性问题,获得合适的聚类中心和高的聚类准确率。

Non-Euclidean Type of Possibilistic C-Means Clustering

The improved possibilistic C-means(IPCM) algorithm is a combination of fuzzy C-means(FCM) algorithm and possibilistic C-means(PCM) algorithm. Based on IPCM and by using the robust statistical viewpoint and the influence function,a new non-Euclidean distance is introduced to replace the Euclidean distance in IPCM.And then a novel fuzzy clustering,called the alternative improved possibilistic Cmeans(AIPCM) clustering model is proposed.The new non-Euclidean distance is more robust than the Euclidean distance.So AIPCM is more robust than IPCM and FCM.Furthermore,with the new non-Euclidean distance,AIPCM can deal with noises or outliers than IPCM and FCM.Numerical simulations demonstrate that AIPCM can overcome the noise sensitivity and obtain more appropriate cluster centers and the improved clustering accuracy.