An effective IDS using CondenseNet and CoAtNet based approach for SDN-IoT environment

A developing technology called the Internet of Things (IoT) allows smart objects to interact over various heterogeneous channels, whether wired or wireless. However, for traditional networks, effectively controlling and managing the data flows of many devices has become difficult. Software-defined networking (SDN) provides a solution to this problem. It has attempted to address several IoT issues, such as flexibility, diversity, and intricacy, because it is programmable, adaptable, fast, and provides a broad overview of the network. As a result, the system for attack detection and mitigation presented in this research leverages deep learning techniques to analyze SDN logs. Once the attack detection process has begun, the input data can be preprocessed using various techniques to replace missing values and prepare the data for additional processing. Subsequently, CondenseNet and Osprey Optimization Algorithm (OOA) are utilized for feature extraction and selection to identify more noteworthy characteristics. Lastly, the very accurate attack prediction is provided by the CoAtNet-based classifier, which is in charge of identifying intrusions. An efficient mitigation procedure was carried out to shield the network from attack after intrusion detection. Furthermore, a conditional tabular generative adversarial network is used to augment the data and correct class imbalance. To validate our proposed model, we conducted research and testing on InSDN, Bot-IoT, and IoT-23 datasets and achieved 99.74 %, 99.61 %, and 99.64 % accuracy, respectively. These experimental findings demonstrate that the suggested framework performs better than current state-of-the-art systems, achieving higher accuracy and a lower false alarm rate.