模糊C均值聚类算法的改进研究

The Research of Improved Fuzzy C-means Algorithm

模糊C均值聚类算法(FCM)是一种比较有代表性的模糊聚类算法,主要是通过迭代更新聚类中心和隶属度矩阵,使目标函数值达到最小.FCM算法还有很多缺陷和不足,其中最主要的就是选取不同的初始中心,会得到不同的聚类结果,影响到聚类的稳定性和准确率.本文对要聚类的数据集采用数据分区技术进行预处理,根据物质质心的定义及质心运动原理,计算每个数据分区的质心做为FCM聚类的初始聚类中心.实验结果表明,改进后的算法FCM能够降低迭代次数和运行时间,得到比较稳定的聚类结果.

The fuzzy C-means algorithm（FCM） is a more representative fuzzy clustering algorithm.It makes the objective function have smallest value by updating cluster centers and membership matrix iteratively.But FCM algorithm has many flaws and shortcomings.It is the most important to select different initial center.It will get different clustering results and affect the stability and accuracy of clustering.In response to these problems,we first want to pre-clustering data sets by data partitioning.And then we calculate the center of each data partition as the FCM＇s initial cluster centers according to the definition of material and the centroid principle of mass motion.Experimental results show the improved FCM algorithm can reduce the number of iterations and running time and get stable clustering results.