MACHINE LEARNING AT THE EDGE: A DATA-DRIVEN ARCHITECTURE WITH APPLICATIONS TO 5G CELLULAR NETWORKS

Abstract

To meet the ultra-low latency demands of future applications, the fifth generation of cellular networks (5G) will rely on edge cloud installations. In this research, we show that similar deployments may also be deployed in mobile networks to allow sophisticated data-driven and Machine Learning (ML) applications. We propose an edge- controller-based cellular network design and assess its performance using real-world data from hundreds of base stations of a large US operator. In this context, we will discuss how to dynamically cluster and associate base stations and controllers based on users' worldwide movement patterns. The controllers will then be used to run ML algorithms to forecast the number of users in each base station, as well as a use case in which these predictions are utilized by a higher-layer application to direct vehicular traffic based on network Key Performance Indicators (KPIs). We demonstrate that prediction accuracy increases when based on machine learning algorithms that depend on the controllers' view and, as a result, on the spatial correlation provided by user mobility, compared to when the prediction is based only on the local data of each individual base station.