基于子空间聚类算法的流量分类方法研究

Network Traffic Classification Method Research Based on Subspace Clustering Algorithm

目前网络流量业务类型具有不断变化和业务特征不断更新两大特点,但是,现有的流量分类器由于存在业务特征库更新代价大、误判率高等缺点,而无法满足正常的业务分类需求。因此需要设计一种子空间聚类算法来实现业务分类精细化,保障分类精确率、召回率以及效率等特性。实验验证表明,子空间聚类算法的业务分类精细化程度高,分类精确率平均超过95%,训练数据需求量低,并且这类方法对于改进DPI分类器对网络环境的适应能力有重大意义。

Currently,service types,features of network traffic are changing constantly,but existing classification methods aren＇t able to satisfy such network traffic environment,because they lack capability to update features library efficiently,and have high misjudgement rate.So a subspace clustering algorithm was designed to test classification properties.Experemnts show that it can classify lots of business types,its classification precision rate exceeds 95%,and quantity demand of training samples is low.It is recommended to help DPI classifier adapt to changing network environment.