Enhancing Intrusion Detection Through Deep Learning and Generative Adversarial Network
Md Habibur Rahman, Leo Martinez, Avdesh Mishra, Mais Nijim, Ayush Vardhan Goyal, David L. Hicks · 2024
Network behavior during intrusion deviates from the standard, which can be identified by establishing a baseline of typical activity. Accurate detection of diverse attack classes with machine learning relies on having adequate representative samples for each class. Typically, highly imbalanced datasets like the NSL-KDD led to a biased model favoring the dominant classes. To address the challenge, we employ a Conditional Tabular Generative Adversarial Network (CTGAN) for generating synthetic samples to balance the dataset effectively. Later on, a deep neural network with a final layer comprising 4 neurons is applied for multi-class classification. The proposed method is compared with state-of-the-art approaches that utilize Conditional Generative Adversarial Network (CGAN) and Wasserstein Conditional Generative Adversarial Network (WCGAN) sampling techniques is found to yield an average improvement of 105.93%, 56.37%, and 80.05% based on Precision, Recall, and F1 Score.