基于流中前 5 个包的在线流量分类特征

针对在线流量分类所面临的特征计算复杂和分类性能不稳定问题，利用流开始的前 5 个数据包（排除三次握手数据包），计算数据包大小、负载大小和到达间隔时间等网络流量的统计特征，通过分析 3 种机器学习算法（C4. 5、BayesNet 和NBTree）分类的结果，研究可用于在线流量分类的特征以及这些特征应该满足的条件。实验结果表明，所提特征计算简单，能快速有效地区分不同的流量，对于不同的机器学习算法，均取得了较高的分类准确率（92%以上），适用于在线流量分类。

Features for Online Traffic Classification Based on the First Five Packets of a Flow

Aiming at the problem of feature computing complexity and instable performance in online traffic classification,features for online traffic classification and its necessary conditions are investigated by analyzing the classification result of three kinds of machine learning algorithms.These features include packet size,payload size,inter-arrival time and other network traffic statistic features which are calculated on the first five packets of a flow that removes three-hand-shake packets.Experimental results show that these features are simple to compute and can discriminate different kinds of traffic quickly and effectively.To different kinds of machine learning algorithms,it can obtain high accuracy(above 92%).These features are suitable to online traffic classification.