基于互信息选择聚类集成的网络流量分类方法

发布时间：2018-03-17 05:18

本文选题：聚类集成　切入点：K均值　出处：《计算机应用》2013年01期 　论文类型：期刊论文

【摘要】：针对互联网流量标注困难以及单个聚类器的泛化能力较弱,提出一种基于互信息(MI)理论的选择聚类集成方法,以提高流量分类的精度。首先计算不同初始簇个数K的K均值聚类结果与训练集中流量协议的真实分布之间的规范化互信息(NMI);然后基于NMI的值来选择用于聚类集成的K均值基聚类器的K值序列;最后采用二次互信息(QMI)的一致函数生成一致聚类结果,并使用一种半监督方法对聚类簇进行标注。通过实验比较了聚类集成方法与单个聚类算法在4个不同测试集上总体分类精度。实验结果表明,聚类集成方法的流量分类总体精度能达到90%。所提方法将聚类集成模型应用到网络流量分类中,提高了流量分类的精度和在不同数据集上的分类稳定性。
[Abstract]:According to Internet traffic marking difficulties and poor generalization ability of single clustering is proposed, based on mutual information (MI) integrated methods clustering theory, in order to improve the traffic classification accuracy. K means clustering results calculated with different initial cluster number K between the true distribution and training flow protocol standardization of each other information (NMI); then based on the NMI values selected for clustering ensemble average K base clustering K sequence; finally using two mutual information (QMI) favorable function to generate consistent clustering results, and the cluster labeling using a semi supervised method. By comparing the clustering ensemble method with a single clustering algorithm in 4 different test sets the overall classification accuracy. The experimental results show that the overall accuracy of traffic classification clustering ensemble method can reach 90%. the proposed method clustering model is applied to network traffic In the classification, the accuracy of the flow classification and the classification stability on different data sets are improved.

【作者单位】：西北工业大学计算机学院;咸阳师范学院信息工程学院;
【基金】：国家863计划项目(2009AA01Z424) 陕西省教育厅专项(12JK0933)
【分类号】：TP181;TP393.06

【参考文献】