Intrusion Detection Method Based on Autoencoder Generative Adversarial Network

Abstract

Intrusion detection system is a proactive security protection technology. Aiming at the data imbalance and unsupervised learning problems in the field of network intrusion detection, AuCoGAN, an intrusion detection model based on autoencoder generative adversarial network, is proposed. The model transforms the generator of Generative Adversarial Networks (GANs) into an encoding-decoding structure, and utilizes the advantage of unsupervised learning of Generative Adversarial Network to build the model by training only the normal network data in the training phase. In the testing phase, the normal network data is passed through the generator, and the difference between the data before and after reconstruction is generally small, while the abnormal data is passed through the generator, and the difference between the data before and after reconstruction is generally large, and the abnormal network data can be distinguished by the size of the Euclidean distance between the data before and after reconstruction. Using the KDD99 dataset for detection, compared with the commonly used traditional machine learning and deep learning models, this model improves in accuracy and detection rate, and reduces the false positive rate.