一种基于量子机制的分类属性数据层次聚类算法  

A hierarchical clustering algorithm of categorical attributive data using quantum mechanism

作　　者：赵正天[1] 赵小强[1] 李炜[1] 段晓燕[2] 卢勇 

机构地区：[1]兰州理工大学电气工程与信息工程学院,甘肃兰州730050 [2]兰州石化职业技术学院电子电气工程系,甘肃兰州730060 [3]中国石油兰州石化电仪事业部,甘肃兰州730060

出　　处：《兰州理工大学学报》2009年第5期89-94,共6页Journal of Lanzhou University of Technology

基　　金：甘肃省自然科学基金(0809RJZA005)

摘　　要：受物理学中量子机制特性的启发,结合层次凝聚思想,通过引入新的相异性度量测度以及聚类度量尺度步长sβtep概念,重新定义以紧致性指标AIAD和离散性指标AIED为基础的聚类有效性函数CVF,提出一种针对分类属性数据的基于量子机制层次聚类算法CQHC.该算法首先在不同粒度水平上划分数据样本产生初始类(簇),然后以聚类有效性函数CVF为评价标准,动态地合并初始类(簇)完成聚类.仿真实验采用2个真实数据集,即:线性可分的大豆疾病样本数据集和线性不可分的动物园数据集.实验结果表明,该算法与已有的其他几个算法相比,不仅具有更高的聚类准确率,而且能够准确地检测出最佳类别数,是有效且可行的.Enlightened by quantum mechanics in physics and incorporated with agglomerative hierarchical clustering, a quantum mechanism-based hierarchical clustering algorithm of categorical attributive data CQHC was proposed by introducing a new dissimilarity measure and a concept of clustering measure scale step βtep, and redefining the cluster validity function CVF based on compactness index AIAD and discrete- ness index AIED. In this algorithm of CQHC, the data sample was partitioned first according to different granularities levels to generate initial clusters. Then the initial clusters were dynamically merged by taking the cluster validity function CVF as evaluation standard and the clustering was completed. Two real data sets, including linear separable soybean disease data sets and linear inseparable zoo data sets, were used for simulation experiment. Experimental result demonstrated that the proposed algorithm was effective and feasible, which not only had higher clustering accuracy, but also accurately detected the best cluster number when compared to other algorithms available.

关 键 词：分类属性 量子机制 层次凝聚 聚类度量尺度步长 聚类有效性函数 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]