Network Intrusion Detection Based on Balancing Generative Adversarial Network

Aiming at the network security problems faced by resource-constrained Internet of Things (IoT) edge devices, this paper proposes an intrusion detection method based on balancing generative adversarial network, namely BAGAN-BD, which uses BAGAN-BD for data enhancement to balance the data distribution, so as to solve the problem that highly unbalanced data will affect the classification detection performance. BAGAN-BD firstly inherits the prior knowledge of the autoencoder with tag embedding, then adopts a multi-generator structure to generate different types of attack data, adds a classifier structure, uses Focal Loss to calculate classification losses, and finally introduces gradient penalty technology to solve the problems of GAN training instability and mode collapse. The experimental results show that the proposed method achieves 97.54% macro average accuracy, 96.90% recall rate and 97.19% F1 score in classification tasks on UNSWNB15 dataset, all of which are higher than the comparison model.