Article: Multidimensional covert traffic attack detection via coupled spatio-temporal transformer and causal convolutional networks Journal: International Journal of Reasoning-based Intelligent Systems (IJRIS) 2025 Vol.17 No.12 pp.35 - 44 Abstract: To address the persistent challenge of detecting traditional model-eluding covert attacks - including low-rate distributed denial of service (DDoS), advanced persistent threat (APT) infiltration, and network steganography - we propose stealth-targeted criss-cross network (ST-CCNet): a multi-dimensional traffic analysis model that integrates spatio-temporal transformer with stacked causal convolutions. The architecture employs causal convolution to extract localised spatio-temporal patterns, while the transformer encoder captures global contextual dependencies. A trainable gated fusion module dynamically synthesises multi-dimensional features (temporal, protocol headers, statistical metrics). Evaluated on the Communications Security Establishment-Canadian Institute for Cybersecurity Intrusion Detection System 2018 (CIC-IDS2018) benchmark, ST-CCNet achieves an improvement of 12 percentage points in recall for stealth attacks (e.g., Slowloris, botnet, web attack) and attains a 98.2% F1-score, outperforming state-of-the-art detectors. This framework provides a robust solution for securing complex network infrastructures against evolving threats. Inderscience Publishers - linking academia, business and industry through research

Title: Multidimensional covert traffic attack detection via coupled spatio-temporal transformer and causal convolutional networks

Authors: Wenji Chi

Addresses: Normal Branch College, Yanbian University, Yanji, 133000, China

Abstract: To address the persistent challenge of detecting traditional model-eluding covert attacks - including low-rate distributed denial of service (DDoS), advanced persistent threat (APT) infiltration, and network steganography - we propose stealth-targeted criss-cross network (ST-CCNet): a multi-dimensional traffic analysis model that integrates spatio-temporal transformer with stacked causal convolutions. The architecture employs causal convolution to extract localised spatio-temporal patterns, while the transformer encoder captures global contextual dependencies. A trainable gated fusion module dynamically synthesises multi-dimensional features (temporal, protocol headers, statistical metrics). Evaluated on the Communications Security Establishment-Canadian Institute for Cybersecurity Intrusion Detection System 2018 (CIC-IDS2018) benchmark, ST-CCNet achieves an improvement of 12 percentage points in recall for stealth attacks (e.g., Slowloris, botnet, web attack) and attains a 98.2% F1-score, outperforming state-of-the-art detectors. This framework provides a robust solution for securing complex network infrastructures against evolving threats.

Keywords: stealth attack detection; spatio-temporal transformer; causal convolution; multi-dimensional traffic analysis.

DOI: 10.1504/IJRIS.2025.150501

International Journal of Reasoning-based Intelligent Systems, 2025 Vol.17 No.12, pp.35 - 44

Received: 13 Jul 2025
Accepted: 09 Oct 2025
Published online: 15 Dec 2025 *

Title: Multidimensional covert traffic attack detection via coupled spatio-temporal transformer and causal convolutional networks

Keep up-to-date