Title: A malicious traffic detection method based on Bayesian meta-learning for few samples

Authors: Zhibin Liu; Zhanpeng Lv; Lixin Zhao; Min Li; Xin Liu

Addresses: North China Branch of State Grid Corporation of China, Beijing, China ' School of Information Science and Engineering, Lanzhou University, Lanzhou, Gansu, China ' North China Branch of State Grid Corporation of China, Beijing, China ' North China Branch of State Grid Corporation of China, Beijing, China ' School of Information Science and Engineering, Lanzhou University, Lanzhou, Gansu, China

Abstract: Realistic network environments have difficulties collecting malicious traffic data, and training network models with virtually generated traffic data are inevitably disconnected from the real network situation. To address few sample problems, we propose a Bayesian meta-learning-based technique to detect encrypted malicious traffic. The internal loop of this meta-learning method is replaced by an analytical marginal likelihood calculation that can be directly implemented as a single optimiser. Experiments show that when the sample size of malicious traffic is reduced to 100, our model still detects up to 96.35%.

Keywords: meta-learning; few samples; cross-domain detection; encrypted traffic.

DOI: 10.1504/IJES.2023.139043

International Journal of Embedded Systems, 2023 Vol.16 No.3, pp.235 - 244

Received: 17 Feb 2023
Accepted: 09 Jul 2023
Published online: 10 Jun 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article