Study on Preliminary Performance of Algorithms for Network Traffic Identification

At present, more and more scholars paid attention to accuracy rate of Internet traffic identification, but not on construction model time, test time, CPU utilization, memory consumption, and concision of models description between different algorithms. However, these indicators played a decisive role in performance of practical internet traffic identification system. So, we collected traffic from existing operatorspsila networks, and tested 15 kinds of supervised learning algorithms and so on. The multivariate evaluation method was proposed to assess the test results. The results show that the 15 kinds of algorithms have similar accuracy rate, but their construction model time, test time, and concision of models description are very different. The C4.5 algorithm is the most suitable classification algorithms for network traffic identification.