Network Traffic Intrusion Detection Strategy Based on E-GraphSAGE and LSTM

The exponential growth of the internet, alongside rapid advancements in information technology, has ushered in an era where network security is critical. As malicious actors employ increasingly complex attack strategies, traditional intrusion detection methods struggle to keep pace. Addressing this gap, we introduce an innovative intrusion detection technique that synergizes Edge Graph Sampling and Aggregation (E-GraphSAGE) with Long Short-Term Memory (LSTM) networks. This approach effectively harnesses both spatial and temporal patterns within network traffic data, enhancing feature retention during training. In addition, with reference to the residual design, we include the features of the network flow itself when aggregating the features to reduce the feature loss caused by the aggregation. Through extensive testing on two benchmark intrusion detection datasets, our method demonstrates superior efficiency in binary and multi-classification tasks, thereby promising a more robust prediction of network traffic behaviors.