The efficiency analysis of the statistical feature in network traffic identification based on BP neural network

How to identify which application generated the traffic attracts many attentions in recent years. Using statistical features can identify the network traffic efficiently without detect the payload of every packet. As the new efficient self-learning algorithm, neural network are imported in to the statistical feature approach. However, constructing the neural network is a high time complexity process and the statistical features used to constructing the neural network are interrelated, which can lead low identification accuracy and the large quantities of computation. So in this paper, in order to filter the more effective statistical features to construct the neural network, we analyze the efficiency of each statistical feature based on the massive experiments. We believe that these results could provide researchers and engineers with important insights for constructing more efficient and accurate neural network to identify the network traffic.