基于机器学习的网络流量分类算法分析研究

Research on Network Traffic Classification Algorithm Based on Machine Learning

基于应用的流量分类在网络安全和管理中具有非常重要的作用。传统流量分类大部分是基于端口的预测方法和基于有效载荷的深度检测方法。由于当前网络环境中各种隐私问题以及基于动态端口和加密的应用,传统的网络流量分类策略的有效性已经逐步下降,目前主要集中在基于机器学习技术的流量分类模型进行研究。本文对各种基于机器学习算法的流量分类的比较,如贝叶斯网络(Bayes Net)、朴素贝叶斯(Naive Bayes)、基于RBF的SVM流量分类和基于遗传算法的SVM(GaSVM)流量分类等。这些算法分别使用了全特征选择和优化后的特征集合,实验结果表明基于遗传算法的SVM流量分类精度较高,并在使用主成分特征也可以达到很高的精度。

Traffic classification based on their generation applications plays an important role in network security and management. The port-based prediction methods and payload-based deep inspection methods comes under traditional methods. The standard strategies in currentnetwork environment suffer from variety of privacy issues,dynamic ports and encrypted applications. Recent research efforts are focused on traffic classification based on Machine Learning Techniques,and made comparison the various Machine Learning（ML） techniques such as Bayes Net,Naive Bayes,SVM based on RBF,VM based on genetic algorithm for IP traffic classification. These classification algorithms used full feature selection and optimized feature set to classify network traffic. It can be seen from the experimental results that GaSVM traffic classification can achieve high accuracy,especially in the use of principal component features.