A Deep Learning-Based Framework for Enhanced Cyberattack Detection and Mitigation in Software Defined Networks

The phenomenal advancements and vehement use of Software Defined Networks (SDNs) require robust cyberattack defense strategies. This arises from traditional intrusion detection systems (IDS), which often fail to solve the intricate security problems posed by SDNs. Defenses are challenged by substantial data volumes and the intricacies of changing network setups, resulting in inadequate attack detection and mitigation. Moreover, while advantageous for network administration, SDN's centralised architecture presents exploitable vulnerabilities. Therefore, to tackle these challenges, this paper introduces an innovative defence mechanism against cyberattacks in SDN. This leverages advanced deep learning techniques to enhance the precision and accuracy of detection and mitigation. This suggested model incorporates an Enhanced CenterNet architecture specifically designed for network traffic, supplemented with knowledge graphs to overcome traditional feature extraction restrictions. In this architecture, to enhance attack classification robustness, a hybrid DenseNet-201 model incorporating network topology, user behavior, and historical attack patterns is implemented. This is further augmented by adversarial training to counter sophisticated attacks. Especially, this dynamic defense mechanism, orchestrated by a remote SDN controller, reconfigures network resources in real-time for a prompt response to distributed denial-of-service (DDoS) attacks. To verify the effectiveness of the proposed model, an experimental analysis has been conducted on InSDN and DDoS-SDN datasets. This analysis is implemented in the Python/Mininet environment. From the results, it is observed that the proposed model achieved significant improvements in precision (4.5%), accuracy (5.9%), recall (4.5%), AUC (2.9%), and specificity (3.9%) of attack detection, reducing response delay by 10.4% when compared to conventional deep reinforcement learning (DRL) and hybrid quantum-classical convolution neural network (HQCNN). Additionally, it improves attack prevention precision (1.9%), accuracy (1.5%), recall (2.5%), AUC (3.5%), and specificity (2.9%), with a 3.5% delay reduction. Thus, this work significantly advances SDN cyberattack defense mechanisms and provides a robust solution to evolving security challenges.