网络流量分类测度的相关性分析1correlationanalysisfor.pdfVIP

网络流量分类测度的相关性分析 1 * 胡晓艳 龚俭 (东南大学计算机科学与工程学院；江苏省计算机网络技术重点实验室 南京 210096) 摘要： 理论上说，网络流量分类测度越多，就能从更多的角度刻画流量以便更精确地分类流量。但过多的 测度使它们之间的关系复杂化，且其测量及处理耗费大量资源。通过对流量分类测度的数据分析后发现， 不少测度之间存在相关性，因此本文提出使用变量聚类，基于测度之间的相关性，将 181 个测度分成了 32 类，即将具有强相关性测度自动聚类，理清了测度之间的关系，并发现了一些潜在的事实，同时找到 了每类测度的代表性测度；然后使用回归分析建立每类的代表性测度与该类中其他测度之间的量化关系， 在确保一定精度的情况下实现测度之间的互推，进而减少部分测度的测量开销。 关键字：网络流量分类测度 相关性分析 变量聚类 回归分析 中图分类号：TP393.0 文献标识码：A Correlation Analysis for Discriminators in Flow Classification XIAOYAN HU, JIAN GONG School of Computer Science Engineering, Southeast University; Key Laboratory of Computer Network Technology in Jiangsu, Nanjing 210096, China Abstract: In theory, the more discriminators for flow classification, the more aspects of flows they would characterize so that the accuracy of classification process would be higher. However, it is difficult to figure out the relationships among so many discriminators and it is expensive in measuring and processing these discriminators. Fortunately, it is found that some discriminators are highly correlated. And thus, in this paper, variable clustering is introduced to clarify relationships by clustering 181 discriminators into 32 clusters based on the correlation between them and selecting a representative discriminator for each cluster. Then regression analysis is used to establish quantified relationship between a cluster representative and other discriminator in the cluster so as to predict the discriminator from the representative and eventually reduce measuring cost. Keywords: discriminators in flow classification, correlation analysis, variable clustering, regression analysis 1. 背景 类测度越多，就能从越多的