A Comparison of Machine and Statistical Time Series Learning for Encrypted Traffic Prediction
Qing He, Georgios P. Koudouridis, György Dán · 2020 International Conference on Computing, Networking and Communications (ICNC) · 2020
This paper studies the utilization of machine and statistical learning methods for predicting encrypted user traffic. To this end, a reference system model and performance metrics for traffic prediction have been defined for enabling on-line training. Based on a collection of representative traffic data sets including various video and web traffic, two different classes of predictors have been evaluated. Our results show that very good prediction performance can be achieved using long short-term memory (LSTM) recurrent neural networks at the price of a significant computational cost for training.