Identification of Network Traffic Using Neural Networks

The solution of the problem of identifying network traffic will allow operators and infrastructure owners to make decisions about the strategy of its service, as well as predict its behavior in the future. This solution will help in the design of virtual or physical networks. In this paper, the problem of identifying several processes (stationary Poisson, MMPP, renewal) is solved using neural networks. First, the classification subtask is solved, which allows to tell which process model the input data correspond to. And, secondly, the subtask of parameter estimation is solved, which allows to tell which values of its parameters better describe the input data. Fully connected and recurrent neural networks are considered as tools. Predictive ability was assessed using classical metrics of regression and classification tasks, as well as using additional metrics.