一种基于Mahalanobis距离的增量聚类算法  

AN INCREMENTAL CLUSTERING ALGORITHM BASED ON MAHALANOBIS DISTANCE

作　　者：郑宏亮[1] 王建英[2] 

机构地区：[1]辽宁师范大学计算机与信息技术学院,辽宁大连116081 [2]辽宁师范大学数学学院,辽宁大连116029

出　　处：《计算机应用与软件》2011年第12期38-40,共3页Computer Applications and Software

基　　金：国家自然科学基金(10771092);辽宁省科技厅博士启动基金(20081079);大连市科学技术基金(2010J21DW019)

摘　　要：经典的模糊c均值聚类算法对非球型或椭球型分布的数据集进行聚类效果较差。将经典的模糊c均值聚类中的欧氏距离用Mahalanobis距离替代,利用Mahalanobis距离的优点,将其用于增量学习中,提出一种基于马氏距离的模糊增量聚类学习算法。实验结果表明该算法能较有效地解决模糊聚类方法中的缺陷,提高了训练精度。Classical fuzzy c-means clustering algorithm is inefficient to cluster non-spherical or elliptical distributed datasets.The paper replaces classical fuzzy c-means clustering Euclidean distance with Mahalanobis distance.It applies Mahalanobis distance to incremental learning for its merits.A Mahalanobis distance based fuzzy incremental clustering learning algorithm is proposed.Experimental results show the algorithm can not only effectively remedy the defect in fuzzy c-means algorithm but also increase training accuracy.

关 键 词：模糊C均值聚类 MAHALANOBIS距离 增量学习 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]