A method of unbalanced twin auxiliary classifiers GAN for network intrusion based mutual information

Abstract

Generative Adversarial Network used in the field of Network Intrusion Detection has become very common, but mode collapse of Generative Adversarial Network and unbalanced distribution of training dataset in Network Intrusion Detection are problems worth solving. Generator and discriminator of Generative Adversarial Network can not fully learn feature information. In this paper, the Twin Auxiliary Classifier GAN is combined with the idea of mutual information modeling. The training is carried out on the Network Intrusion Detection dataset UNSW-NB15. After comparing the original dataset trained by Naïve Bayes, Decision Tree, Random Forest, Support Vector Machine and Multi-layer Perceptron Machine with the expanded dataset generated by Twin Auxiliary Classifier GAN training, the results show that the methods proposed in this paper can improve the performance of each classifier on the test set.