Article: Network traffic anomaly detection driven by bidirectional self-attention mechanism Journal: International Journal of Information and Communication Technology (IJICT) 2026 Vol.27 No.54 pp.65 - 84 Abstract: In the face of increasingly covert cyber-attacks, traditional detection models struggle to effectively capture the complex contextual correlation features in the traffic, resulting in insufficient ability to identify new threats. To address this issue, this study proposes a detection model based on bidirectional self-attention mechanism, which achieves deep perception of abnormal behaviours by simultaneously learning the context information of the traffic sequence. Experimental results show that compared with mainstream long short-term memory and standard transformer methods, this model has an average area under the curve improvement of over 4.2%, and the recall rate for low-rate attacks has increased by 7.5%, significantly enhancing the accuracy and robustness of detection. This study provides a new idea for improving the active defence capability of network security. Inderscience Publishers - linking academia, business and industry through research

Title: Network traffic anomaly detection driven by bidirectional self-attention mechanism

Authors: Jing Luo; Jie Song

Addresses: Department of Computer Science and Technology, Sichuan Police College, Luzhou, 646000, China ' Department of Computer Science and Technology, Sichuan Police College, Luzhou, 646000, China

Abstract: In the face of increasingly covert cyber-attacks, traditional detection models struggle to effectively capture the complex contextual correlation features in the traffic, resulting in insufficient ability to identify new threats. To address this issue, this study proposes a detection model based on bidirectional self-attention mechanism, which achieves deep perception of abnormal behaviours by simultaneously learning the context information of the traffic sequence. Experimental results show that compared with mainstream long short-term memory and standard transformer methods, this model has an average area under the curve improvement of over 4.2%, and the recall rate for low-rate attacks has increased by 7.5%, significantly enhancing the accuracy and robustness of detection. This study provides a new idea for improving the active defence capability of network security.

Keywords: cybersecurity; anomaly detection; self-attention; bidirectional encoding.

DOI: 10.1504/IJICT.2026.153719

International Journal of Information and Communication Technology, 2026 Vol.27 No.54, pp.65 - 84

Received: 26 Jan 2026
Accepted: 27 Feb 2026
Published online: 21 May 2026 *

Title: Network traffic anomaly detection driven by bidirectional self-attention mechanism

Keep up-to-date