基于半监督学习的P2P协议识别

P2P classification using semi-supervised learning

随着P2P技术的发展,网络上充满了大量的P2P应用。协议加密技术的发展,使得P2P应用的识别和管理变得非常困难。描述了如何运用半监督的机器学习理论,根据传输层的特征,用聚类算法训练数据并建立一个高效的在线协议识别器,用于在内核协议层对协议特别是P2P协议进行识别,并对BitComet和Emule进行了实验,得到了很高的识别准确率(80%)。研究并解决了将选取好的特征用于聚类并高效地实现最后的协议识别器。

P2P classification is important for the ISP services and network administrators these years, as the fluent new P2P applications.The traditional identification skills such as port based and payload based technologies can＇t classify any protocols now, especially the encrypted protocols. A semi-supervised clustering based classification model is used for P2P application identification. The classification accuracy and the system＇s activity is mainly discussed. The feature selection and online classification are what the paper concentrates on.