CycleGAN-based intrusion detection data augmentation model

A model is proposed to address the problem of imbalanced sample data classes in response to attack behavior in intrusion detection systems based on deep learning technology, using an optimized CycleGAN (Cyclic Consistency Adversarial Network) to expand the sample of intrusion detection data. Firstly, the original dataset is divided into data categories and the adversarial generation network GAN is trained. Then, based on CycleGAN, a double-layer adversarial generation network is used to establish dual constraints on it to obtain new sample data, which can alleviate the impact of data class imbalance on classification tasks and enhance the key feature representation of the generated samples. The experimental results conducted on the KDDCUP99 dataset showed that using raw data and enhanced sample data to train the DNN detection classifier, the average recognition rate for U2R and R2L was significantly improved after using the enhanced dataset, and the overall average recognition accuracy was also slightly improved.