A study of supervised machine learning algorithms for traffic prediction in SD-WAN

Modern cloud, web and other emerging distributed services have complex network requirements that cannot be fulfilled via classical networks. This paper presents a novel architecture of a noble software-defined wide area network (SD-WAN) that provides the framework for incorporating AI/ML based components for managing different centralised services of the WAN. To leverage the benefit of this framework, a crucial early stage requirement is to accurately identify the traffic category of a flow based on which follow-up actions such as QoS provisioning, resource orchestration, etc. can be implemented. To address this, the research then presents the model of a supervised ML based traffic prediction module and presents a detailed comparison and performance analysis of a shortlisted set of ML models with a variety of traffic categories. The research also takes into account the serialised processes in the models' training and learning phases emphasising on the sensitivity of the feature selection process in the performance of these algorithms.