Data Augmentation of Discrete Sequential Protocol Messages Based on Recurrent Generative Adversarial Networks
Chang Zhao, Qing Li, Xintai He, Runze Wang, Kun Chen, Zeying Liu · 2022 2nd International Conference on Consumer Electronics and Computer Engineering (ICCECE) · 2022
In some special networks such as IoT, industrial control networks, and sensor networks used in various industries, data are usually discrete and short, called Discrete Sequential Protocol Messages (DSM). DSM data contain less information, so feature extraction is difficult, and the effect of classification is poor. For imbalanced DSM data, the problem is more complex. This paper proposes a data augmentation algorithm based on Recurrent Generative Adversarial Networks (RGAN), and uses Long short-term Memory Network (LSTM) to improve the generator structure, making the model more suitable for DSM data. Under the multiple conditions of data imbalance, the binary classification and multi-classification tasks are carried out on 2012 MACCDC and USTC-TFC 2016 datasets respectively. The results show the RGAN method we proposed can solve the problem of data imbalance under various imbalanced conditions. It can effectively improve the classification performance of minority classes and ensure the overall classification performance. For the data length, the RGAN method can process imbalanced DSM data of more than 10 bytes, which shows RGAN applies to short data.