GBERT: a hybrid BERT and GRU model for enhanced encrypted traffic classification

As encrypted traffic becomes increasingly prevalent in modern network communications, effective classification of encrypted traffic is crucial for network security and management. However, existing classification methods face significant challenges when dealing with encrypted traffic, especially in capturing the complex temporal features and contextual information of traffic data. For this reason, this study proposes the GBERT model, a novel hybrid neural network structure that integrates the contextual understanding capabilities of the BERT model with the temporal processing advantages of Gated Recurrent Unit (GRU). Validated across several public encrypted traffic datasets, the GBERT model demonstrates exceptional performance, notably achieving classification accuracies of 99.03% on the ISCX-VPN-Service dataset and 99.32% on the USTC-TFC dataset. Additionally, this study explores various GRU configurations, offering valuable insights for selecting the most effective model architecture for specific tasks.