Abnormal User Behavior Generation based on DCGAN in Zero Trust Network
Yansheng Qu, Jianfei Chen, Ming Li, Yunxiao Wang, Ning Li, Hua Huang, Bo Mao · Procedia Computer Science · 2022
It is essential to detect the abnormal user behavior in zero trust network. Currently, the deep learning based models have been applied to the user detection, and these models requires huge volume of labeled data for the training. However, the abnormal data is not common in the network. Therefore, it is necessary to create a program to simulate the abnormal user behavior and to generate the training data. In this paper, a GAN is trained to create the network attacks for the training of detection deep learning models. The user behavior is first converted to an attribute image in which one dimension represents the type of network behavior and another dimension represents the time. Therefore, the generated user behavior not only includes the behavior but also the temporal information and frequency. Also an improved DCGAN model is applied. The experimental results indicate that the proposed method can be used to generate the user behavior data for the training of abnormal detection model. Compared the baseline GAN model, the proposed method can improve the overall accuracy of the attack data generation by 7%.