Machine Learning Methods for Traffic Prediction in Dynamic Optical Networks with Service Chains

Knowledge about future traffic in a dynamic optical network can be used to improve various performance metrics, including network cost and to reduce complexity of solving network optimization problems. In this paper, we propose a machine learning approach of predicting demands in a dynamic optical network serving Virtual Network Function (VNF) chain traffic. We also present numerical results proving effectiveness of the described methodology and showing comparison of various classifiers. Datasets for experiments were generated based on real network topology.