RISKS OF USING NETWORK INTRUSION DETECTION SYSTEMS AS A SOURCE OF TRAINING LABELS FOR NEURAL NETWORKS
DOI:
https://doi.org/10.34185/1991-7848.itmm.2026.01.069Keywords:
NIDS, machine learning, cybersecurity, PU-learning, concept drift, error self-amplification, online model adaptationAbstract
The use of network intrusion detection system (NIDS) alerts as training labels for machine learning models introduces systematic biases that significantly degrade detection accuracy. This study investigates the discrepancies between actual network attacks and signature-based triggers, focusing on three critical challenges: one-sided labeling bias, error self-amplification in continuous learning environments, and vulnerability to adversarial data poisoning. Specifically, the inability of traditional NIDS to identify zero-day threats results in a polluted negative class, where missed attacks are misclassified as legitimate traffic. To address these risks, mitigation strategies are analyzed, including positive-unlabeled (PU) learning, weak supervision, and confidence-based filtering mechanisms. Implementing these robust validation protocols and buffering techniques ensures more reliable threat detection and enhances the resilience of neural networks against evolving cyber threats in dynamic network environments.
References
Feng Y., Sakurai K. Network Intrusion Detection: Evolution from Conventional Approaches to LLM Collaboration and Emerging Risks. URL: https://arxiv.org/abs/2510.23313.
Dilworth R., Gudla C. Applications of Positive Unlabeled (PU) and Negative Unlabeled (NU) Learning in Cybersecurity. URL: https://arxiv.org/abs/2412.06203.
Caravan: practical online learning of in-network ML models with labeling agents / Q. Zhang et al. 18th USENIX symposium on operating systems design and implementation (OSDI 24). Santa Clara, CA, 2024. P. 325–345. URL: https://www.usenix.org/conference/osdi24/presentation/zhang-qizheng.
Zou H. P., Caragea C. JointMatch: a unified approach for diverse and collaborative pseudo-labeling to semi-supervised text classification. Proceedings of the 2023 conference on empirical methods in natural language processing / ed. by H. Bouamor, J. Pino, K. Bali. Singapore, 2023. P. 7290–7301. URL: https://doi.org/10.18653/v1/2023.emnlp-main.451.
Alajaji A. FortiNIDS: defending smart city iot infrastructures against transferable adversarial poisoning in machine learning-based intrusion detection systems. Sensors. 2025. Vol. 25, no. 19. P. 6056. URL: https://doi.org/10.3390/s25196056 (date of access: 04.03.2026).
Managing Concept Drift in Online Intrusion Detection Systems with Active Learning / C. F. et al. URL: https://www.tib.eu/de/suchen/id/base:1b52787437b97f11f6c3a39a28994f83fc750b5f.
Sommer R., Paxson V. Outside the closed world: on using machine learning for network intrusion detection. 2010 IEEE symposium on security and privacy. 2010. P. 305–316. URL: https://doi.org/10.1109/SP.2010.25.
