Network Intrusion Detection Method Based on CGAN and CNN-BiLSTM

To a certain extent, the Internet has substantially enriched people's life due to its convenience. At the same time, cyber-attacks are common and pose a significant threat to personal privacy and public safety. Real-time detection and warning of network behavior and traffic are crucial for network security. A network intrusion detection (NID) method, known as CGAN-CBiLSTM, was developed to adapt to the requirements of the current network environment. In our algorithm, the conditional generative adversarial network (CGAN) was combined with the convolutional neural network (CNN) and bi-directional long short-term memory neural network (BiLSTM). In data preprocessing, raw traffic data was mapped into the gaussian domain. CGAN can generate virtual samples based on specified labels to build balanced traffic datasets. In the hybrid neural network, CNN and BiLSTM were mainly responsible for extracting the temporal and spatial features of the traffic. Experiments on the NSL-KDD dataset verified the proposed method, and the classification accuracy, precision, f1-score, and FAR can achieve 88.25%, 96.61%, 88.7%, and 3.7%, respectively. Our algorithm considered the data imbalance problem and designed a suitable model structure. As a result, CGAN-CBiLSTM achieved high accuracy in the NID task.