Design of Network Security Intrusion Detection System Based on NIDSeqFormer Framework

In the constantly evolving field of network security, the effectiveness of Network Intrusion Detection Systems (NIDS) is crucial. This study proposes a network security intrusion detection system based on the NIDSeqFormer framework, which is a modular machine learning pipeline designed to handle sequence to sequence models, in response to the traditional Transformer model's neglect of the serial nature of network communication in identifying the long-term behavior and characteristics of rich networks.It consists of four main components: Pre Processing,Input Encoding, Model, and Classification Head, making it suitable not only for standard tasks but also for handling more complex sequence analysis tasks.To highlight the usefulness and efficiency of the NIDSeqFormer framework, we conducted detailed studies on three frequently used public datasets. We report the accuracy, speed, and model size findings. It was found in the experiment that the selection of input encoding and classification heads has a significant impact on the performance of themodel. After system testing experiments, the results showed that by selecting specific input codes and classification heads, the accuracy of the model can reach over 97%,and the inference and training time can be improved without any loss of accuracy.