摘要
准确的网络流量分类是一系列网络管理活动的重要基础。近年来,利用机器学习的原理处理流量分类问题成为网络测量领域的研究热点,其中朴素贝叶斯方法因分类速度快,实现简单等优点而被广泛应用,但随着网络流量的变化和网络类别的增加,该方法的分类准确性和鲁棒性随着时间增长而逐渐降低,引入了一种新的流量分类模型更新方法,通过对模型的更新训练提升其分类性能,并保证模型应用的稳定性。理论分析和实验结果都表明:该方法能够使流量分类模型随着时间的增长而保持良好的总体性能,且不易受报文抽样的影响,能为其他诸多网络活动提供相应的支持。
Accurate network traffic classification is an important base for a series of network management activities. In recent years, the traffic classification using the theory of machine learning has been a hot topic in network measurement. Being simple and efficient, the Naive Bayes method has been widely applied, however, with the changes of network flows and the increases of network classes, the accuracy and robustness of this method are decreased. A new update approach for classification model was proposed, improving the performance of traffic classification by update the original model and ensuring the stability of the model application. Theoretical analysis and experimental results show that: this method enables traffic classification model to maintain good overall performance as the growth over time, and few is affected by packet sampling and to provide appropriate support to many other network activities.
出处
《系统仿真学报》
CAS
CSCD
北大核心
2013年第11期2597-2603,共7页
Journal of System Simulation
基金
国家973计划(2009CB320505)
国家科技支撑计划(2008BAH37B04)
关键词
流量分类
朴素贝叶斯
统计特征
报文抽样
traffic classification
naivebayes
statistical feature
packet sampling