A network traffic classification and anomaly detection method based on parallel cross-convolutional neural networks

Network traffic anomaly detection, an effective means of network defence, can detect unknown attack behaviours and provide crucial support for network situation awareness. However, existing methods face challenges such as reliance on manually designed features, decreased classification accuracy, slow processing speeds, and loss of important information in traffic. To solve these problems, inspired by the binocular vision principle, we propose a parallel cross-convolutional neural network model. The model directly extracts original network traffic payload data as input, controlling depth. Utilising two deep convolutional neural network (CNN) data transformation streams undergoing three cross-blends, more feature information is extracted, enabling the capture of deeper traffic characteristics. Experimental results on the USTC-TFC2016 dataset demonstrate our model achieves 100% accuracy with only two epochs for 20-class classification, outperforming other similar models in detection performance. © 2024 Inderscience Enterprises Ltd.