An Incremental Learning Model for Network Intrusion Detection Systems
Shang-Te Wang, Shi-Sheng Sun · 2024
With the ever-increasing complexity of Internet security issues, new attack types can be detected every day. The Network Intrusion Detection System (NIDS) can quickly identify and deal with attacks. In this paper, we establish an intrusion detection model suitable for NIDS through incremental learning which can continuously learn the behaviors from new network flows. We utilize the UNSW_NB15 network intrusion dataset, and a portion of the data is used to train the initial model, while the remaining data simulates new network flows. Our proposed model can continuously learn new network flow patterns and adjust parameters without storing original data, while improving the model's training time and maintaining accuracy.