基于模糊最大散度差判别准则的自适应特征提取模糊聚类算法

摘要

The derivation mistake of clustering center and the related wrong conclusion in Gao' s fuzzy maximum scatter difference discriminant criterion based clustering algorithm (FMSDCA) are pointed out. A new clustering algorithm based on fuzzy maximum scatter difference discrinminat criterion (FMSDC) ,called as fuzzy compactness and sepation clustering algorithm based on fuzzy maximum scatter difference discriminant criterion (FMSDC-FCS),is proposed. FMSDC-FCS make use of the FMSDC to generate optimal projection vector and make use of the fuzzy compactness and separation (FCS) algorithm to cluster the reduced-dimensional data set. The projection vector and clustering result are optimized by alternately running FMSDC in the original data space and FCS in the projection space, and the original data is clustered by clustering the reduced-dimensional data. The experimental results demonstrate that the overall performance of FMSDC-FCS surpasses that of original FCS algorithm, FMSDCA and classical fuzzy c-means algorithm.%指出皋军等人提出的基于模糊最大散度差判别准则(Fuzzy Maximum Scatter Difference Discriminant Criterion,FMSDC)的聚类算法(Fuzzy Maximum Scatter Difference Diseriminant Criterion Based Clustering Algorithm,FMSDCA)中聚类中心表达式的推导错误及相关结论的错误,在修改该错误的基础上提出新的基于FMSDC的模糊聚类算法:FMSDC-FCS (Fuzzy Compactness and Separation Clustering Algorithm Based on Fuzzy Maximum Scatter Difference Discriminant Criterion).FMSDC-FCS利用FMSDC产生最佳投影矢量,利用模糊紧性分离性(Fuzzy Compactness and Separation,FCS)算法对降维数据聚类,通过交替运行原数据空间中的FMSDC和投影空间中的FCS来优化投影矢量和聚类结果,最终通过对降维数据的聚类实现对原始数据的聚类.实验结果表明,FMSDC-FCS总体性能优于原有的FCS算法、FMSDCA算法以及经典的模糊C-均值算法.