一种改进的模糊C均值聚类算法

A Novel FCM Clustering Algorithm

针对经典的C均值聚类算法以及模糊C均值聚类算法所存在的两个方面的问题:一是算法对初始聚类中心的过分依赖性,通常的聚类算法往往对于不同的初始聚类中心会得到不同的聚类结果;二是算法需要预先知道实际的聚类数目,而在实际应用中,聚类数目却是未知的。基于此提出了模糊C均值聚类算法的一种改进算法,即在标准的模糊C均值聚类算法的基础上,给目标函数加入了一个惩罚项,使得上述问题得以解决。并通过仿真实验证实了新算法的可行性和有效性。

There are two issues in the application of FCM clustering algorithm： one is that the FCM algorithm is too sensitive to the initial cluster centers, people can get different clustering result from different original clustering center, and the other is that the number of the clusters C needs to be determined in advance as an input to the algorithm, but C always does not be known. Based on this, a novel algorithrn of FCM is proposed in this paper. Bassed on the FCM, a penalty term is added into the objective function and the above - mentioned issues can be resolved. The simulation demonstrates the feasibility and validity of the proposed method.