works against adversarial domain name system tunnuling attacks using hybrid neural networks

The increasing prevalence of adversarial attacks targeting Domain Name System (DNS) infrastructures has raised significant concerns regarding network security. DNS tunneling, a technique exploited by attackers to bypass security mechanisms, can potentially lead to unauthorized data exfiltration and network compromise. This study proposes a novel approach to secure networks against adversarial DNS tunneling attacks using a hybrid neural network model, specifically a Voting Classifier, which combines the predictions of multiple base classifiers to improve attack detection accuracy. The model is trained on the FGSM_combined.csv dataset, which includes both adversarial and non-adversarial DNS traffic, sourced from the Kaggle "Adversarial Machine Learning" dataset. By leveraging the strengths of ensemble learning techniques and deep learning models, the proposed hybrid approach achieves enhanced robustness and performance in identifying DNS tunneling attempts. Experimental results demonstrate that the Voting Classifier outperforms individual machine learning models, providing a promising solution for real-time detection and mitigation of DNS tunneling attacks, thereby contributing to the strengthening of network security frameworks.

   Keywords: Adversarial regional attacks, DNS tunneling, network security, hybrid neural network models, voting classifier, ensemble learning, deep learning, intrusion detection, adversarial machine learning, data exfiltration, RTD mechanism, and security mechanism.