Abstract: Accurate identification and categorization of multimedia traffic is the premise of end to end QoS （Quality of Service） guarantees. Today, the dramatic growth of data volume takes challenge to network traffic classification. Therefore, using feature selection methods for multimedia flow identification and classification is particularly important in the big data era. This paper introduces related works on feature selection using identification and categorization of multimedia traffic, which is divided into three categories：Filter, Wrapper, Embedded and analyzes the performance of these methods. Then, this paper compares the performance of various feature selection algorithms using latest dataset from three aspects：The running speed, the feature compression rate and the feature selection accuracy. Besides, to improve classification accuracy, this paper proposes a hierarchical structure to reach fine-grained classification, according to the dataset.

Key words: features selection, video traffic classification, hierarchical classifier

摘要： 准确，高效的业务流识别与分类是保障多媒体通信端到端QoS（Quality of Service），执行相关网络操作的前提。如今数据规模的剧烈增加为业务流的分类提出了挑战，而特征选择能够尽可能地减少特征维数，去除冗余特征，为大数据时代下的业务流分类提供解决办法。对现有的特征选择方法分成Filter、Wrapper、Embedded三类，分析了各类算法的性能原理。采用最新数据集对不同特征选择算法性能对比，从算法的运行时间、特征压缩率、准确率三个方面评估了特征选择算法的性能。另外，针对现有数据集分类情况进行分级分类以达到视频流的细分类，从而提高分类的准确率。

关键词: 特征选择, 视频流分类, 多级分类器