Cyberattack defense mechanism using deep learning techniques in software-defined networks

Software-defined networking (SDN) is a network architecture. It is becoming more popular due to its centralized network administration, adaptability, and speed. However, the centralized structure of SDN architecture makes assaults more prevalent. The attacks affect normal users by draining server resources, reducing internet speed, and occupying memory on controllers and switches. Therefore, security for SDN is essential. Most of the existing attack detection methods for SDN are based on statistical and machine learning-based techniques. In statistical techniques, determining an accurate threshold is difficult due to dynamic nature of the network flow. For machine learning-based techniques, it can be difficult to identify a suitable feature that can distinguish assaults from regular traffic. Therefore, this work presents an effective deep learning-based framework to identify network threats in SDN. This framework comprises a data augmentation generative adversarial network (DAGAN), Xception, and improved ShuffleNetV2 models. First, DAGAN is used to increase data samples and reduce class imbalance problem in the dataset. Then, Xception network extracts the essential features, and finally, intrusions are identified and categorized using an improved ShuffleNetV2 network. When a network intrusion is discovered, the suggested defense mechanism is turned on to restore normal network connectivity quickly. Several tests are carried out on two SDN-based datasets, and our proposed approach surpasses existing models by achieving 89.63% and 98.96% accuracy for InSDN and Ton-IoT datasets, respectively. Additionally, our suggested model delivers a fair trade-off between recall and precision that qualifies it for attack categorization.