A Brute-force Black-box Method to Attack Machine Learning-Based Systems in Cybersecurity

Machine learning algorithms are widely utilized in cybersecurity. However, recent studies show that machine learning algorithms are vulnerable to adversarial examples. This poses new threats to the security-critical applications in cybersecurity. Currently, there is still a short of study on adversarial examples in the domain of cybersecurity. In this paper, we propose a new method known as the brute-force attack method to better evaluate the robustness of the machine learning classifiers in cybersecurity against adversarial examples. The proposed method, which works in a black-box way and covers some shortages of the existing adversarial attack methods based on generative adversarial networks, is simple to implement and only needs the output of the target classifiers to generate adversarial examples. To have a comprehensive evaluation of the attack performance of the proposed method, we use our method to generate adversarial examples against the common machine learning based security systems in cybersecurity including host intrusion detection systems, Android malware detection systems, and network intrusion detection systems. We compare the attack performance of the proposed method against these security systems with that of state-of-the-art adversarial attack methods based on generative adversarial networks. The preliminary experimental results show that the proposed method, which is more efficient in computation and outperforms the state-of-the-art attack methods based on generative adversarial networks, can be used to evaluate the robustness of various machine learning based systems in cybersecurity against adversarial examples.