基于代价敏感决策树的网络流量分类研究  

Research on Network Traffic Classification Based on Cost-sensitive Decision Tree

作　　者：吴耿[1] 李杰[1] 杨文保[1] 

机构地区：[1]中南大学信息科学与工程学院,湖南长沙410075

出　　处：《电脑与信息技术》2011年第5期5-8,共4页Computer and Information Technology

摘　　要：传统的基于端口的流量分类方法和基于DPI技术的流量分类方法由于P2P技术和加密技术的流行而开始失效。基于网络流特征及机器学习的流量分类方法因为克服了上述弊端而成为了流量分类领域的研究热点。实际网络环境中,"大象流"和"老鼠流"在数量和传输字节量等方面存在着严重的不平衡,降低了基于机器学习流量分类方法的实际分类效果。针对该问题,文章将代价敏感决策树C4.5_cs算法应用于网络流量的分类当中。实验证明,文章采用的方法具有更高的"字节分类准确率",适合于不平衡网络流量的分类。Traditional traffic classification method based on port number and DPI technologies are becoming ineffective as that P2P and encrypt technology becoming popular. Because traffic classification method based on network flow statistics and machine learning can resolve prior defect, it becomes a hot in traffic classification field. In practical network environment, there are serious imbalance in quantity and bytes carried between ＂elephant flows＂ and ＂mouse flows＂, which impacts the actual classification result of machine learning traffic classification method. To resolve this problem, this paper uses the cost-sensitive decision tree to the traffic classification. Experimental results show that our method has greater ＂bytes classification accuracy＂ and suitable to the classification of imbalance network traffic.

关 键 词：网络流量分类 不平衡流量 代价敏感 决策树 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]