摘要
阐述了一种无监督连接划分聚类算法,算法基本思想是首先通过分割的方法将数据集划分为若干个原子簇,滤除噪声原子簇,然后通过对原子簇间连接亲密度的分析,构造原子簇间的连接图,切断连接亲密度很低的原子簇连接,合并连接亲密度高的连接,划分得到最后的聚类结果。算法具有很高的有效性,适用于高维数据集,能够对任意形状的簇进行聚类。通过分析与实验,证明该方法具有良好的效果。
A kind of unsupervised clustering algorithm based on link division was presented. First the algorithm uses partitioning method to divide data set into several atomic clusters and gets rid of noise atomic clusters. Then by analyzing the consanguinity among atomic clusters, it transfers to link graph, cut links of low consanguinity and comb high ones. Finally, result of clustering could be got. With high effectiveness, the algorithm could be performed in high dimensional data set and an arbitrary shape of clusters. Experiment shows that this approach is well available for use.
出处
《计算机工程与设计》
CSCD
北大核心
2006年第3期384-386,共3页
Computer Engineering and Design
基金
教育部科学技术研究基础条件平台建设基金项目(505006)
关键词
无监督聚类
连接划分
任意形状聚类
入侵检测
网络安全
unsupervised clustering
link division
arbitrary shape clustering
intrusion detection
network security
