Crosslayer network outage classification using machine learning

Abstract

Network failures are common, difficult to troubleshoot, and small operators with limited resources need better tools for troubleshooting. In this paper, we analyse two years of outages from a small global network for high-quality services. Then, we develop a machine learning model for outage classification that can be set up with little effort and low risk. We use passive Bidirectional Forwarding Detection (BFD) data to classify Layer2 problems and add active packet loss data to classify other problems. The Layer2 problems were classified with a 99% accuracy and the other problems with 40%--100% accuracy. This is a significant improvement when we observe that only 35% of the customer cases we studied received any Reason for Outage (RFO) response from the Customer Support Centre.