Combining Random Forest and Linear Regression to Improve Network Traffic Prediction

The constant increase in traffic volume triggers the development of new network optimization methods. Machine learning techniques are one of the promising research directions, as they provide the necessary tools to improve network performance in various aspects. Furthermore, prediction-based algorithms are getting the attention of researchers to achieve more precise network operations. In this paper, we approach the traffic prediction problem in backbone optical networks. We propose a model based on the Voting Regressor, combining Random Forest and Linear Regression. We create two large semi-synthetic datasets with daily patterns of various popular network-based services and applications. Through broad numerical experiments, we show how the proposed approach outperforms the reference ones in different traffic profiles.