Multi-Class Network Traffic Generators and Classifiers Based on Neural Networks
Radion Bikmukhamedov, Adel Nadeev · 2021 Systems of Signals Generating and Processing in the Field of on Board Communications · 2021
We introduce a neural network framework that allows constructing multi-class network traffic models suitable for flow generation and classification tasks. Packet size and inter-packet time sequences were the only flow features that formed the inputs. Our approach has two main components: a packet feature quantizer and a sequence model, which we verified on example configuration with K-Means algorithm and Transformer-based models respectively. Moreover, we claim a possibility of knowledge transfer from the generation to the classification task. The evaluation of the generator's traffic quality was carried out by means of Kolmogorov-Smirnov metric for such parameters as packet size, inter-arrival time, throughput and number of packets per flow in directions to and from traffic origin. The results showed that the proposed multi-class model performed on par with individual single-class Markov-based models. As for the classification task, the model outperformed FS-NET and a conventional Random Forest classifier. We demonstrated that performing the knowledge transfer procedure between traffic generator and classifier (so-called generative pretraining) could improve performance on the traffic classification task even further. We also release the code, datasets and trained models.