Identification of Unknown Malicious Flow Based on Adaptive Annotation and Deep Neural Networks

With the explosive growth of Internet flow, the problems of network security are becoming more and more prominent. The identification and detection of malicious flow in the network have become important topics. Most of the existing classification models are based on closed-world data sets, and can only classify the sampled flow data. In this case, unknown flow, especially the hidden malicious flow, cannot be effectively identified. In order to solve this problem, this paper proposes a recognition framework based on deep neural network to identify unknown flow and detect malicious flow, which can automatically expand the known data set to enhance the classification and detection ability of the model. Firstly, the framework constructs a classification model that can effectively identify the network flow based on the convolutional neural network. Based on this model, the unknown flow data with potential security risks are distinguished and clustered. Then, the unknown flow is maliciously detected according to the clustering results, and the model is updated with the newly discovered flow data.