在基于机器学习方法的网络流量分类系统中,特征产生及特征选择将直接影响到分类的速度及精度。针对这种情况,在特征产生方面,结合Packet-Level和Flow-Level上的信息,对报文属性(大小、个数、时间、标志位)、流的属性(时间)进行分析和研究,产生了37个网络流量统计特征。在特征选择方面,提出一种结合Filter模型和Wrapper模型的网络流量特征选择方法,实现了降维作用。实验表明,提出的方法提高了分类的精度。 In the network traffic classification system based on machine learning method, feature generation and feature selection will directly affect the classification speed and accuracy. In this case, packet attributes (size, number, time, flag) and attribute (time) of packets are analyzed and researched based on the information of Packet-Level and Flow-Level in terms of feature generation. 37 network traffic statistics. In the aspect of feature selection, a method of network traffic feature selection based on Filter model and Wrapper model is proposed, which realizes dimensionality reduction. Experiments show that the proposed method improves the classification accuracy.