A C-IFGSM Based Adversarial Approach for Deep Learning Based Intrusion Detection

Abstract

With the rapid development of machine learning algorithms, the security problem has gradually emerged. Most existing algorithms may be attacked by adversarial examples. An adversarial example is a slightly modified input sample that can lead to a false result of machine learning algorithms. This poses a potential security threat for many machine learning-based applications. Especially in the domain of intrusion detection, the intrusion adversarial examples may result in malicious attacks on intrusion detection classifiers. To our knowledge, all previous work only apply the adversarial examples generation methods in the field of image classification, which is not suitable for network traffic datasets. Aiming at generating more similar intrusion adversarial examples, this paper explores a Constraint-Iteration Fast Gradient Sign Method (C-IFGSM) that can adapt to complex network traffic datasets with multiple types of features and multiple relationship among features. Experiments show that the C-IFGSM based adversarial approach can achieve good performance on intrusion adversarial examples.