Adversarial Attacks Against Black-Box Network Intrusion Detection Based on Heuristic Algorithm

Machine learning has become an increasingly utilized approach for developing Network Intrusion Detection Systems (NIDS) for identifying anomalies associated with malicious network intrusions. Even in a black-box attack setting, where the attacker only has access to the output, machine learning-based classifiers are vulnerable to adversarial examples. This paper proposes a universal perturbation generation method based on a heuristic algorithm for generating adversarial examples to perform black-box attacks against different technology based NIDSs. The simulation results show that the adversarial examples generated by the proposed perturbation generation approach reduce the accuracy of multiple single and hybrid NIDS models.