Securing Networks Against Adversarial Domain Name System Tunneling Attacks Using Hybrid Neural Networks
Stephanie Ness · IEEE Access · 2025
Domain name system tunneling is one of the emerging threats that use Domain name system (DNS) to transfer unwanted material, and it is usually undetected by conventional detection systems. Thus, the current paper proposes a double-architecture deep learning system built upon Long short-term memory (LSTM) and Deep Neural Networks (DNN) to detect and categorize adversarial Domain name system tunneling assaults. Limitations in the current Domain name system traffic classification techniques are overcome in the proposed model through temporal sequence modelling and feature extraction to distinguish clearly between normal, attack, and adversarial traffic. Based on the experiments conducted on a broad data set, the application of the proposed hybrid model increased the classification accuracy up to 85.2%, which is higher compared with basic machine learning algorithms. Moreover, the ablation analysis showed that downstream components, such as the Long short-term memory layer and exact dropout rate, are critical to the performance of the proposed model against adversarial perturbation. This work offers a solution for identifying intricate threats in a big and live manner; as such, it has broad applicability in sensitive areas of activity like finance, health care, and administration. Further work includes applying this approach to other network-based threats and improving the effectiveness of applying it to oligopolistic adversaries’ tactics.