Using machine learning for dynamic multicast capacity planning

Abstract

We are using service consumption data from BT’s UK–wide IPTV service to identify main drivers of network capacity and to predict changes on the level of exchanges. We have used a decision tree to identify main drivers and find that provisioning data is sufficient to identify capacity requirements for cable links to exchanges. We have used cluster analysis to identify changes in service consumption and to construct an early warning system for potential capacity bottlenecks.