一种基于属性分解的FCM融合聚类算法

摘要

针对传统模糊C均值聚类算法只能发现“类球状”簇和对分量属性数据敏感的缺点,提出一种基于FCM的属性分解聚类再融合的分类算法.该算法将信息融合的思想应用于聚类算法,先在每个分量属性维度进行聚类,然后对各属性的聚类结果进行融合分析并得到聚类结果.独立对每个分量属性聚类的思想为算法的并行实现提供便利.实验结果表明,该算法不但能有效提高聚类的准确度,而且不需要提前对数据进行归一化处理,在分量属性量测数据存在偏差时仍然表现出良好的鲁棒性.%Aiming at the deficiencies of traditional fuzzy c-means (FCM) clustering that it can only find ball-like clusters and is sensitive to attribute data of components,a classification algorithm of FCM-based attribute decomposition clustering re-fusion (FFBAD) is proposed in this paper.This algorithm introduces the idea of information fusion to clustering algorithm,makes clustering on attribute dimension of every component first,and then fuses and analyses the clustering results of each attribute to yield the clustering result.The idea of clustering the attribute of each component independently makes it convenient in parallel implementation of algorithm.Experimental results show that this algorithm can improve clustering accuracy,and does not need to perform normalised processing on data in advance.Moreover,it can also demonstrate good robustness when the deviation exists in measuring data of component attribute.