ICMPv6-Based DoS and DDoS Attacks Detection Using Machine Learning Techniques, Open Challenges, and Blockchain Applicability: A Review

Although the launch of Internet Protocol version six (IPv6) addressed the issue of IPv4's address depletion, but also mandated the use of Internet Control Message Protocol version six (ICMPv6) messages in newly introduced features such as the Neighbor Discovery Protocol (NDP). This has exacerbated existing network attacks including ICMPv6-based Denial of Service (DoS) attacks and its variant form Distributed Denial of Service (DDoS) attack. Intrusion Detection Systems (IDS) aimed at tackling security issues raised by ICMPv6-based DoS and DDoS attacks have been reviewed by researchers and a general classification of existing IDSs was proposed as anomaly-based and signature-based. However, it is incredibly hard to see the overall picture of IDSs based on Machine Learning (ML) techniques with such a classification, as there is a lack of a more detailed view of the ML approach, classifiers, feature selection techniques, datasets, and different evaluation metrics. Nevertheless, recent developments in this relatively new field have not been covered such as ML-based IDSs using flow-based traffic representation. Therefore, this article specifically reviews and classifies IDSs based on ML techniques to detect ICMPv6-based DoS and DDoS attacks as single and hybrid classifiers. In addition, blockchain applicability in Collaborative IDS (CIDS) architecture based on the ensemble framework has been proposed as a solution to one of the open challenges for ICMPv6-based DoS and DDoS attacks detection problem. Moreover, this review also provides a classification of ICMPv6 vulnerabilities to DoS and DDoS attacks which would provide a reference resource for future researchers in this domain. To the best of the author's knowledge, this is the first review paper specifically focusing on IDSs based on ML techniques in this domain, as well as blockchain applicability as a possible research direction has been proposed to attract researcher's focus on building ensemble learning-based IDS models.