Open-Set Intrusion Detection Based on Adversarial Reciprocal Points Learning and Manifold Mixup

As network technology continues to advance rapidly, cybersecurity threats are evolving at an unprecedented pace, posing significant challenges to traditional intrusion detection methods. In this context, promptly detecting unknown network attacks is essential for reducing potential damage and strengthening security defenses. To tackle this challenge, this paper presents an open-set intrusion detection approach leveraging adversarial reciprocal points learning and manifold mixup. The approach aims to accurately identify known threats while effectively detecting previously unseen attacks, thereby strengthening network security. The proposed method leverages Convolutional Neural Networks (CNNs) to extract network traffic features and integrates adversarial reciprocal points learning to capture attack characteristics not encountered during training. Furthermore, manifold mixup is used to create pseudo unknown attack samples, enhancing the model’s capacity to identify new threats. Experimental results on the CICIDS2017 dataset show that the proposed approach achieves a recall rate of 89.34% for unknown attacks and an AUROC score of 95.18%. These results confirm the method’s effectiveness in tackling cybersecurity issues in open environments, significantly boosting intrusion detection performance.