By Ilana Weinstein, Comcast Ramya Narayanaswamy, Comcast Federica Mutti, Comcast Aaron Tomkins, Comcast
Broadband access network continues to experience substantial growth in High-Speed Data (HSD) capacity demands year over year, with customers expecting 100% availability all the time. Internet service is viewed as an essential service, like electric power and gas. All multiple system operators (MSOs) face the daunting challenge of minimizing customer service disruptions and staying ahead of potential issues to detect and mitigate before customers notice the issue.
In this paper we cover an unsupervised learning approach to cluster all relevant features and help in obtaining directionality toward potentially multiple areas that may be contributing to an anomaly. This directionality toward interpretable areas will serve as a good starting point for any manual investigation needed and will aid operations teams and subject matter experts in identifying the source of the problem with the end goal of reducing customer disruption and enabling a positive customer experience.