Abstract: Traditionally, in the area of the network traffic classification, there exists a problem that single learning algorithm lacks classification accuracy and is incapable of adapting to the dynamic network environment. Accordingly, it proposes a novel classification approach which is a combination of multi-classifier. This method combines the features of a range of classifiers and then achieves traffic classification by means of majority voting and instance selection. Moreover, comparative experiments show that this method improves the classification accuracy, the generalization performance and the ability to adapt to the dynamic network environment. However, it is worth noting that the method has a larger implement complexity and time complexity than these of single algorithm.

Key words: traffic classification, Support Vector Machine（SVM）, C4.5 decision tree, Bayesian Net（BN）, ensemble learning

摘要： 针对流量分类问题中，传统单一的机器学习分类算法存在分类准确率难以提升和对网络环境变化适应能力不足的缺点，提出一种多分类器集成流量分类方法。该方法结合不同算法分类器的特点，使用多数投票和实例选择集成方法实现流量分类。对比实验表明，该方法在分类准确率和算法泛化性能上的表现均有所提升，对环境变化适应能力增强。但值得注意的是，该算法比独立分类法从实现复杂度和实际运行的时间复杂度均有所增加。

关键词: 流量分类, 支持向量机, C4.5决策树, 贝叶斯网, 集成学习