Transfer-learning-based Network Traffic Automatic Generation Framework

Nowadays, there is an increasing number of attacks against the network system. The intrusion detection system is a standard method to prevent attack. In essence intrusion detection is a classification problem to judge normal or abnormal behaviors according to network traffic characteristics, and deep learning has been applied to intrusion detection recently. However, due to the lack of training data in some systems such as industrial control systems and smart grid, the deep-learning algorithm cannot give full play to its advantages. To solve this problem, we propose a transfer-learning-based network flow generation framework for deep-learning-based intrusion detection, which uses invariant extraction and sequence to sequence generation, to extract the attack invariant of the existing attack data set and transfer the knowledge to the target network system. We use the open-source data set on real systems and carry out relevant experiments, proving that our method can generate effective anomaly traffic as well as improve the accuracy of intrusion detection.