The Study Of VoIP Traffic Identification Technology Based On Machine Learning

VoIP is a communication model based on the Internet, which can transfer voice, video and text information of user communication. VoIP is different from traditional phones, and it is better than traditional phone. Firstly, it provides a diversity of services. The traditional telephone is limited to the provision of voice services. In addition to the provision of voice services, VoIP also provides video and text transmission services. As well as providing video conferencing services for multi-user. Secondly, it provides more convenient and efficient service than traditional phone. With the popularity of the Internet, the rapid growth of wireless services coverage area, users will be free from the restrictions of time and space to use the VoIP services. What’s more, VoIP service is much cheaper compared to traditional telephone service. Finally, it’s stronger in expandability. Users can communicate with each other and with the traditional telephone communication during different VoIP protocol standard developed applications. These advantages promote the rapid development of VoIP. With the growth of users and the emergence of management issues, the basic of effective management of VoIP is detection and identification of traffic and distinguishing normal traffic from illegal traffic.This paper briefly describes the protocol standards and technology of VoIP and elaborates on the importance of traffic identification. Through analyzing the current method of several VoIP traffic identification, the methods are mainly based on the analyzing the characteristics of the host and flow behavior or the analyzing transport protocol of VoIP. On this basis, this paper proposes a traffic identification method based on machine learning. The method extracts known streaming flow characterized in246alternative feature set, and selects the core feature vector for the Bayesian network training. After that, to achieve the purpose of VoIP traffic identification by making use of Bayesian network training the raw traffic samples, and combining with feature vector. It is to get the normal traffic of VoIP and other software, as well distinguishing the illegal traffic such as DoS attack traffic. Through experimental data analysis and comparison with other traffic identification method, it shows that the indicators such as accuracy, precision and recall is perfect, which means that the method is able to identify the VoIP traffic accurately. Finally, to analyze the relationship between the number of input feature VoIP traffic identification method accuracy in order to achieve the minimum input for maximum accuracy, while reducing the identification method run time complexity and space complexity.With the rapid development of cloud computing and other communication technologies, the network traffic generated by exponential, as well the real-time monitoring was in necessary. The proposed method provides good example for big data and real-time monitoring.