基于数据约减的聚类有效性分析  

Cluster validity analysis based on data reduction

作　　者：于晓[1] 李晨[1] 王亚茹[1] YU Xiao LIChen WANG Ya-ru(School of Electrical Engineering and Automation, Tianjin University, Tianjin 300072, Chin)

机构地区：[1]天津大学电气与自动化工程学院,天津300072

出　　处：《传感器与微系统》2017年第3期55-57,61,共4页Transducer and Microsystem Technologies

基　　金：国家自然科学基金资助项目(61573251)

摘　　要：聚类分析中利用有效性指标判断数据集的正确类数极易受到噪声数据、类之间分离性以及聚类算法的影响,所确定类数的正确性难以得到保证。为克服这个问题,以文献[1]中的数据约减方法为基础,对原数据集和约减后的数据集利用有效性指标进行正确类数判别。实验表明:该方法能增大类之间的分离性,有效判断数据集的最优类数。Estimating the correct number of clusters by cluster validity index in cluster analysis is highly susceptible to noise data,separation among clusters and clustering algorithm,so the correctness of the estimated number of clusters is difficult to be guaranteed. In order to overcome this problem,validity index is used to estimated number of clusters in original dataset and reduced dataset based on the data reducing method proposed in reference[1 ],the result demonstrate the method can enhance separation among clusters and effectively determine the optimal number of clusters.

关 键 词：数据约减 方向角 聚类分析 最优类数 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]