Abstract: In a traffic classification approach based on C4.5 algorithm, the great number of network traffic and its diversity of features make construction speed and classification speed become important criterions for network traffic classification. This paper provides an improved way to calculate the infogain and entropy. It reduces the computation complexity and speeds up the building time of decision tree. Based on the experimental data, this approach demonstrates that it achieves a satisfying accuracy and reduces the building time of classification system at the same time.

Key words: traffic classification, C4.5 algorithm, entropy of information, information gain ratio, decision tree

摘要： 在基于C4.5算法的网络流量分类方法中，网络流量数据量的海量性及其特征的多样性使得决策树的构建速度、分类速度成为评价网络流量分类器的重要标准。在原C4.5算法的基础上提出一种改进的信息熵的计算方法，通过减少计算函数的复杂度，提高决策树的构建速度。实验表明，基于改进后算法的分类器在达到原有分类准确率的同时，极大地缩短了决策树的构成时间。

关键词: 网络流量分类, C4.5算法, 信息熵, 信息增益率, 决策树