GHM-DenseNet Intrusion Detection Method Based on GAN

With the continuous development of computer and Internet in recent years, the situation of network security is not optimistic. Solving and maintaining various problems in network security is an important research content in recent years. Aiming at the low detection rate of a few intrusion like data samples in unbalanced intrusion detection data processed by traditional machine learning methods and current deep learning methods, this paper proposes a method combining Generative adversarial network GAN, GHM Loss, DenseNet and attention mechanism. After pre-processing the original intrusion detection data, the Generative adversarial network is used to over-sample the original data set, expand the scarce data set, and send the expanded data set to DenseNet-based for classification and processing. GHM Loss is used to improve the loss function in the original DenseNet. In order to solve the problem that Focal Loss focuses too much on difficult samples, and the convergence model still fails to judge outliers. At the same time, the dual-attention mechanism is used to make the model pay more attention to important features of data and suppress unnecessary features, so that the model has a higher recognition rate for scarce data sets. Through experiments on NSL-KDD data sets, compared with traditional machine learning and current mainstream deep learning methods, the accuracy and F1 values of the model in this paper are improved, which verifies the validity of the model.