Zhipeng Li, Shaobo Liu, Hao Yang, Rui Dong, Shuo Shen, and Yu Zhang
Network intrusion attacks, secret sharing, abnormal trafficmonitoring, GCN, GAN
With the escalating demand for advanced network technologies and heightened concerns over information security, the detection of anomalous network traffic has become increasingly critical. However, conventional detection methodologies exhibit inherent limitations in handling complex topologies, imbalanced data distributions, and multimodal traffic patterns. To surmount these challenges, we introduce GAGNNet, a novel framework that synergises reinforcement learning with GraphGAN, graph convolutional networks (GCNs), and deep Q-Networks (DQNs) to enhance anomaly detection efficacy. In GAGNNet, GraphGAN is first employed to generate augmented data, effectively expanding the pool of anomalous traffic samples and bolstering the model’s capacity to identify attacks with minimal labelled instances. Subsequently, GCNs are leveraged to comprehensively model traffic topology, capturing intricate structural dependencies among traffic nodes for precise classification. Finally, DQN-based reinforcement learning is incorporated to dynamically optimise the parameters of the GCN model, enabling the detection system to adapt to diverse attack patterns and enhance overall performance. Empirical evaluations conducted on multiple public datasets and real-world network traffic data substantiate the superiority of GAGNNet over conventional machine learning (ML) and deep learning (DL) approaches. The proposed framework demonstrates significant improvements across key performance metrics, including accuracy, precision, recall, and F1-score, underscoring the efficacy and robustness of graph-based methodologies in anomaly detection. This work not only presents an efficient, intelligent, and scalable solution for network security but also provides critical technical support for intelligent traffic monitoring and proactive network security management.
Important Links:
Go Back
