Network Traffic Threat Feature Recognition Based on a Convolutional Neural Network

This research proposes a novel algorithm for identifying and classifying threats from network traffic features based on a Convolution Neural Network (CNN). The CNN extracts higher-dimensional and more abstract network features from the network traffic by deepening the number of convolution layers. This research is inspired by AlexNet, and the network layers of the AlexNet are modified to adapt recognition of the network features, which is expected to guarantee the accuracy while reducing most of the parameters of the CNN structure. This research uses the transfer learning method for data training, which trains the CNN initially with a small amount of dataset, and then trains the network with the complete dataset. This training method improves the memory status of the CNN, and hence it speedup the training process. The presented network threat identification system would successfully be able to distinguish the various threats from the given network traffic features.