Abstract
A robust Deep Reinforcement Learning-based Intrusion Detection Scheme (DRL-IDS) for Software-Defined Networking (SDN) which combines the Long-Short Term Sequence Recurrent Neural Network (LFTS-RNN) with the Particle Cloud-Integrated Joint Time- and Feature-Optimization Algorithm (PC-JTFOA). The hybrid model aims to enhance the security of SDN through the detection and mitigation of a wide array of Distributed Denial of Service attacks and network misbehaviors across different SDN planes. The LFTS-RNN is used for accurate attack detection and misbehavior identification. Meanwhile, the PC-JTFOA optimizes feature selection, load balancing, and energy-efficient routing, thus ensuring fast and reliable network traffic management. The deep reinforcement learning approach further enables continuous adaptation to changing network behaviors, thus making the model dynamically adapt to known as well as emerging attack vectors. The proposed DRL-IDS scheme obtains superior performance in experimental results based on the NSL-KDD and WPPD datasets. The LFTS-RNN model indicates a highly impressive sensitivity of 98.67% and specificity of 97.42%, while the DRL-IDS model presents an detection accuracy of 99.85%. The PC-JTFOA further improves the solution by exhibiting a low response time of 1423 ms, which indicates tremendous improvement in computational efficiency. A comparative analysis with the existent intrusion detection methods pointed out that the scheme proposed not only outperforms other models in terms of detection accuracy as well as adaptability, but it also reduces complexity.