Optimizing DDoS Attack Classification in SDN Networking via Combining Features

Software Defined Networking has evolved as a transformative networking paradigm, enabling the dynamic programmability of network devices and redefining traditional network architecture. This paper addresses the imperative need for effective SDN security, particularly in countering Distributed Denial of Service attacks, which exploit the inherent flexibility of SDN to inflict widespread network disruption. The core goal of this research is to leverage machine learning classifiers - namely Decision Tree, Naive-Bayes, K Nearest Neighbors, and Random Tree - to discern and classify malicious and normal packets within an SDN dataset. To enhance the accuracy of classification, a comprehensive feature selection process is executed, encompassing four distinct techniques: Information gain, gain ratio, ReliefF, and symmetrical uncertainty. The pivotal contribution of this paper lies in the refinement of attribute selection through a combining approach, pinpointing the most salient attributes from the previously identified set. Additionally, a contemporary SDN dataset is employed, specifically tailored to identify distributed denial of service attacks within an SDN environment. The results demonstrate that the combined approach outperforms other methods with an exceptional accuracy of 99.9979% with the Decision Tree algorithm, emphasizing the practical applicability of this methodology. The study employs a diverse array of evaluation criteria, encompassing Precision, Recall, F-measure, and Accuracy, as indicators of performance, to evaluate the model's effectiveness. This research underscores the critical significance of SDN security and showcases the effectiveness of machine learning in countering DDoS threats within SDN networks. Through the innovative fusion of feature selection techniques and advanced classifiers, it delivers substantial enhancements to SDN security, promising robust protection against DDoS attacks within SDN environments.