Dcgan Data Balancing to Improve Accuracy of Hybrid CNN-LSTM Intrusion Detection Framework in SDN Environment

Maintaining robust network security in Software-Defined Networking (SDN) systems has become increasingly challenging due to sophisticated cyber-attacks and the centralized nature of SDN. This paper introduces a novel intrusion detection system based on a hybrid deep learning model that combines Convolutional Neural Networks (CNN) for spatial feature extraction and Long Short-Term Memory (LSTM) networks for temporal dependency extraction. The approach is applied to the large InSDN dataset, having labeled traffic for normal activity as well as various classes of attacks, to train multi-class as well as binary classifiers. Synthetic samples are generated based on Deep Convolutional Generative Adversarial Networks (DCGAN) in order to effectively tackle the issues due to class imbalance and thereby enhance the detection rate for minority classes of attacks. Experimental tests carried out in a simulated SDN network with Mininet and Hping3 have outstanding performance, with the binary model achieving 99.81% accuracy and the optimal multi-class model achieving 99.4% accuracy. Such promising results demonstrate the capability of the proposed framework to offer an efficient and scalable real-time intrusion detection solution for the modern SDN infrastructures.