改进的基于概化的概念构成聚类算法  

Improved Generality-based Concept Formation Clustering Algorithm

作　　者：甘丽[1] 

机构地区：[1]唐山学院计算机科学与技术系,河北唐山063000

出　　处：《煤炭技术》2011年第3期182-184,共3页Coal Technology

摘　　要：基于概化的概念构成(Generality-based Concept Formation,GCF)是一种分类数据层次聚类算法,对GCF算法提出2点改进。首先,定义了一种新的基于条件概率分布的相似度度量,并用它替代原算法中的相似度,该度量将分类数据进行数值化处理,更精确地反映了数据间的相似程度。其次,提出相似度品质概念,给出了计算公式,相似度品质可与原算法中样本变异系数配合使用,共同确定概化水平。改进算法提高了聚类准确率,同时算法的时间复杂性保持不变。Generality-based Concept Formation（GCF） is a type of hierarchical clustering algorithm for categorical（symbolic） data.Improvements on GCF have been made in two aspects as follows.Firstly,a novel categorical similarity metric based on conditional probability distribution is present and used in the improved algorithm.Not only can the similarity metric measure the difference in the numbers of dissimilar attributes between two clusters,but also the key lies in expressing the different degrees of them,whose nature is to do numerical processing of categorical data.Secondly,this paper suggests the notion of similarity quality and designs its calculation formula.In the improved algorithm,the similarity quality is utilized with the sample variation coefficient at the same time so as to render the generality level used in the clustering process to be more adaptive.Theoretical analysis and experimental results indicate that the improved GCF algorithm is valid.

关 键 词：GCF 聚类 相似度度量 分类数据 

分 类 号：TP274[自动化与计算机技术—检测技术与自动化装置]