摘要
可能性聚类有两大缺陷:一致聚类中心问题和有效性指标失效问题。对于第一个问题,有人提出在目标函数中添加聚类中心排斥项,但这样会引入更多的参数。为此,本文提出了一种改进的可能性聚类算法,较好地解决了这个问题。对于第二个问题,本文通过对隶属度作适当变换,使修正的有效性指标适用于可能性聚类。实验结果表明,该算法的优越性明显,有效性指标估计更为准确。
Possibilistic clustering has two weaknesses. On the one hand, it always leads to a single cluster center; on the other hand, the existing validity indexes under a fuzzy clustering environment are not workable in possibilistic clustering models. For the first weakness, this paper proposes an improved possibilistic clustering algorithm with a mutual repulsion of the clusters, whose parameters can easily be handled. For the second weakness, this paper has the existing validity index redefined by replacing the possibilistic c-membership with the modified possibilistic c-membership, so that it is workable under a possibilistic clustering environment. Experimental results show that the improved algorithm has better performance, and the validity index works well.
出处
《计算机工程与科学》
CSCD
北大核心
2009年第8期49-51,共3页
Computer Engineering & Science
关键词
可能性聚类
一致聚类中心
有效性指标
possibilistic clustering
identical clustering center
validity index
