一种改进的加权模糊核聚类算法  

Improved Algorithm for Weighted Fuzzy Kernel Clustering Analysis

作　　者：高翠芳[1] 吴小俊[1] 

机构地区：[1]江南大学物联网工程学院,无锡214122

出　　处：《数据采集与处理》2010年第5期631-636,共6页Journal of Data Acquisition and Processing

基　　金：教育部新世纪优秀人才计划(NCET-06-0487)资助项目;国家自然科学基金(60472060;60572034)资助项目;江苏省自然科学基金(BK2006081)资助项目;江南大学创新团队计划(JNIRT0702)资助项目

摘　　要：聚类算法对初始值的依赖性较大,通常容易陷入局部最优,很难得到稳定的聚类结果。为改善该问题,本文提出了一种改进的加权模糊核聚类算法,将迭代自组织的数据分析算法(ISODATA)的思想引入到加权模糊核聚类算法(WFKCA)中,利用聚类中心分裂/合并的中间结果来调整初始中心,降低了WFKCA算法收敛于局部最优的可能。改进算法采用特征空间中的计算度量,将单值标准差阈值扩展为标准差阈值向量,并增加了对聚类中心的调整幅度。实验结果显示,该算法在不同结构和维数的数据集上都取得了更稳定的聚类精度。The dependence of clustering algorithm on initial values is generally liable to stick to a local optimum,therefore makes it difficult to obtain a stable clustering result.To overcome this shortcoming,an improved algorithm for weighted fuzzy kernel clustering analysis is proposed.The idea of iterative self-organizing data analysis technique algorithm（ISODATA） is introduced into the weighted fuzzy kernel clustering algorithm（WFKCA）,and initial center vectors are adjusted by the intermediate results from splitting and/or merging of clustering centers to reduce the possibility of local optimum.The algorithm uses a matchable measurement from the feature space,extends the single-value standard deviation threshold to standard deviation threshold vector,and increases the adjustment range of clustering centers.Simulation results show that the algorithm can achieve more stable clustering accuracy on the benchmark data sets.

关 键 词：模糊核聚类 ISODATA算法 局部最优 稳定性 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]