Metaparameter optimized hybrid deep learning model for next generation cybersecurity in software defined networking environment

The Software Defined Networking (SDN) method has evolved to project future systems and collect novel application needs for several years. SDN delivers sources for enhancing management and system control by splitting data and control plane, and the control logic is federal in a controller. Conversely, the central logical control is a perfect objective for malicious assaults, chiefly Distributed Denial of Service (DDoS) threats. Deep Learning (DL) is one of the influential models useful in cyber-security, and numerous Network Intrusion Detection (NIDS) were developed in current studies. Some researchers have specified that deep neural networks (DNN) subtly perceive adversarial assaults. These attacks are examples of definite worries that cause DNNs to misclassify. Therefore, this manuscript develops a novel Cybersecurity in Software-Defined Networking utilizing Hybrid Deep Learning Models and a Binary Narwhal Optimizer (CSSDN-HDLBNO) approach. The presented CSSDN-HDLBNO approach provides a scalable and effective solution to safeguard against evolving cyber threats in DDoS attacks within the SDN environment. Initially, the CSSDN-HDLBNO approach utilizes min-max normalization to scale the features within a uniform range using data normalization. Furthermore, the binary narwhal optimizer (BNO)-based feature selection is accomplished to classify the most related features. For the DDoS attack classification process, the attention mechanism with convolutional neural network and bidirectional gated recurrent units (CNN-BiGRU-AM) is employed. To ensure optimal performance of the CNN-BiGRU-AM model, hyperparameter tuning is performed by utilizing the seagull optimization algorithm (SOA) model to enhance the efficiency and robustness of the detection system. A wide range of simulation analyses is implemented to certify the improved performance of the CSSDN-HDLBNO technique under the DDoS SDN dataset. The performance validation of the CSSDN-HDLBNO technique portrayed a superior accuracy value of 99.40% over existing models in diverse evaluation measures.