一种基于半监督学习的应用层流量分类方法

基于应用层的流量分类在用户行为识别、网络带宽管理等方面有着十分重要的应用.将机器学习应用到应用层流量分类问题中.首先提出了一种基于熵函数的组合式特征选择算法,提取了5种TCP连接的特征.针对监督学习中无法识别新流量类型的问题,提出了一种基于半监督学习的流量分类算法.实验结果表明,算法的检测率优于Kmeans方法.在少量标记样本的情况下,随着未标记样本数增加,算法的检测率在增加.

Network Application Classi fication Method Based on Semi-Supervised Learning

Categorizing network traffic by application type is challenging because of the continued evolution of applications. In this paper, a feature selection method based on entropy is proposed firstly. Five TCP connection features are selected for classifier. Then, a semi-supervised classification method that allows classifiers to be designed from training data consisting of only a few labelled and many unlabelled is proposed. The experiment shows that the detection results precedes on Kmeans method.