Distributed Denial of Service Classification for Software-Defined Networking Using Grammatical Evolution

Software-Defined Networking (SDN) stands as a pivotal paradigm in network implementation, exerting a profound influence on the trajectory of technological advancement. The critical role of security within SDN cannot be overstated, with distributed denial of service (DDoS) emerging as a particularly disruptive threat, capable of causing large-scale disruptions. DDoS operates by generating malicious traffic that mimics normal network activity, leading to service disruptions. It becomes imperative to deploy mechanisms capable of distinguishing between benign and malicious traffic, serving as the initial line of defense against DDoS challenges. In addressing this concern, we propose the utilization of traffic classification as a foundational strategy for combatting DDoS. By categorizing traffic into malicious and normal streams, we establish a crucial first step in the development of effective DDoS mitigation strategies. The deleterious effects of DDoS extend to the point of potentially overwhelming networked servers, resulting in service failures and SDN server downtimes. To investigate and address this issue, our research employs a dataset encompassing both benign and malicious traffic within the SDN environment. A set of 23 features is harnessed for classification purposes, forming the basis for a comprehensive analysis and the development of robust defense mechanisms against DDoS in SDN. Initially, we compare GenClass with three common classification methods, namely the Bayes, K-Nearest Neighbours (KNN), and Random Forest methods. The proposed solution improves the average class error, demonstrating 6.58% error as opposed to the Bayes method error of 32.59%, KNN error of 18.45%, and Random Forest error of 30.70%. Moreover, we utilize classification procedures based on three methods based on grammatical evolution, which are applied to the aforementioned data. In particular, in terms of average class error, GenClass exhibits 6.58%, while NNC and FC2GEN exhibit average class errors of 12.51% and 15.86%, respectively.