Abstract
Aim/Background: Forthcoming IPv6-based networks will face a momentous security challenge with ICMPv6 communications. An attacker uses ICMPv6 messages to steep the target system and aims to contrivance a Denial of Service (DoS)or Distributed Denial of Service (DDoS)attack. A robust intrusion detection system is being developed by researchers to address these issues. Researchers have access to a restricted number of IPv6 datasets to construct well-known intrusion detection systems. however, these datasets are not accessible to the public and only target on one kind of attack.
Methods: In this study, we primarily concentrate on the development of a benchmark dataset that is labeled and reflects ICMPv6 traffic for intrusion detection systems that focus on DoS/DDoS assaults under IPv6. Our dataset is raised using VMware Workstation Pro and Graphical Network Simulation 3 (GNS3). The attacks are generated by using THC Toolkit and both normal and attack packets are captured by using Wireshark. The dataset is named as IDOS6 (Icmpv6 Based DDoS attack on IPv6). Even though IDOS6 contains the data to evade the icmpv6 based DDoS attack, it could not be gifted to find the Zero-day attacks. Hence our research work further delves into incorporating Generative AI models to generate adversarial DoS/DDoS data samples (AIDoS6) that resemble the real-world traffic data.
Results: According to the experimental results, with the use of the developed datasets, machine learning classifiers like Support Vector Machine (SVM), Random Forest, Decision Tree, MLP, KNN, and Logistic Regression were trained and evaluated using performance metrics like Accuracy, Precision, Recall and F1 Score. SVM and Logistic Regression achieved an accuracy rate of 85%, for IDoS6 and 77.6 % of accuracy for AIDoS6 which is comparatively high when compared to the other machine learning classifiers.
Conclusion: The experiments clearly states that the IDoS6 and AIDoS6 datasets are able to dodge from machine learning and deep learning detection models and share attack characteristics with genuine samples.