SMAD: Robust DGA detection approach based on Improved Attention Mechanism
Published in Journal of Information Security and Applications (JISA), 2026
Domain Generation Algorithms (DGAs) are techniques used by malware to algorithmically generate large numbers of domain names for establishing command-and-control (C&C) communication. We propose SMAD, a DGA detection approach designed to improve robustness against adversarially crafted domains and generalization to previously unseen DGA families. SMAD introduces a Shift-Attention mechanism that models character interactions beyond local dependencies, enabling the capture of meaningful structural features within a global context. Experimental results show that SMAD achieves up to 96% detection accuracy in open-world scenarios and significantly outperforms state-of-the-art baselines in adversarial detection with an AUC of 0.72, while sustaining over 500 QPS with millisecond-level inference latency in real-world deployment.
Recommended citation: Shaoxuan Yun, Yuchao Zhang, and Qianrui Yuan. (2026). "SMAD: Robust DGA detection approach based on Improved Attention Mechanism." Journal of Information Security and Applications. DOI: 10.1016/j.jisa.2026.104488
Download Paper
