Generative Adversarial Networks for Imbalanced Dataset Intrusion Detection in Software-Defined Networking

Software-defined networks (SDN) have become prominent technologies in recent times owing to their centralized network management, flexibility, and rapidity. The centralized structure of SDN architecture may introduce vulnerability and threat, which can affect normal users through resource depletion, decreased internet speeds, and memory consumption on controllers and switches. An efficient intrusion detection system (IDS) is required for actively monitoring and identifying malicious activities or potential threats within SDN networks. The current machine learning techniques in IDS often face challenges when dealing with imbalanced datasets. These datasets can lead to biased model performance toward the dominant class, causing inadequate detection of minority-class instances like anomalies or intrusions. Moreover, a large number of features in the dataset increases computational challenges and may adversely affect the model's performance. This work presents a deep learning-based technique generative adversarial networks (GAN) to generate synthetic data for balancing the imbalanced dataset issues in IDS. This helps in improving detection performance, especially for minority classes. The chi-square test based on statistics is also used to select the most significant features that enhance model performance and decrease both training and testing time. We evaluate the model's performance using multiple machine learning algorithms, including Naive Bayes (NB), Extra Trees (ET), Random Forest (RF), and XGBoost (XGB). Our evaluation demonstrates improved accuracy and reduced training and testing times across these algorithms. Notably, XGB achieves the highest accuracy <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{0. 9 9}$</tex>.