An Online Network Traffic Classification Method Based on Deep Learning

Traffic classification plays an important role in network traffic analysis. In this paper, an online network traffic classification method based on deep learning was proposed. By pre-training CNN and fine-tuning the fully connected layer, the complexity of model update is greatly reduced, thus enabling online learning. To deal with the stability-plasticity dilemma, we introduce a proficiency mechanism. The proposed method uses the raw binary data of packet header as input of the model instead of statistical features, which avoid complex feature selection process. And it can identify traffic by a small number of packets, which meets the needs of early stage identification. Besides, our model has the ability of online learning, and can adapt dynamically to the network environment. Experimental results show good performance of the proposed method in both accuracy and speed on traffic classification task.