BiRNNs-SAT for Detecting BGP Traffic Anomalies in Communication Networks
Zhenxiang He, Chunwei Li, Xian-sheng WANG · 2023
Border Gateway Protocol (BGP) serves as a path vector protocol that manages network reachability information among Autonomous Systems (AS), is critical to the stability and reliability of the Internet. The single approach in previous studies fails to adequately capture the temporal and feature information in BGP traffic data. In this study, we introduce a novel bidirectional recurrent neural networks with self-attentive mechanism (BiRNNs-SAT) method to better detect anomalous behavior in BGP traffic data. In our proposed method, normalization is used to balance the temporal and feature dimensions of the singularity data, while a two-layer hierarchical structure is constructed. The first layer aims to capture bidirectional time-dependent and feature correlation information to provide a comprehensive data representation for the model. The second layer employs a self-attention mechanism to compute the degree of contribution of each hidden state to the attention and dynamically generate the weights between connections. To validate the feasibility of the model, we use five real-world collected datasets for extensive experimental evaluation. The findings show that our suggested BiRNNs-SAT method performs well on the BGP anomaly classification task, improving the F1 value by up to 16.8892% relative to the baseline model. In summary, the BiRNNs-SAT model proposed in this study provides an efficient and effective solution to the BGP anomaly classification problem.