As more and more data is collected from networks, the introduction of machine learning can help MSOs optimize and automate their network responses, speakers told attendees at SCTE’s Cable Tec Expo on Monday.
According to Applied Broadband CEO Jason Schnitzer, the implementation of machine learning algorithms can help operators better optimize their services by separating populations into similar groups. For example, Schnitzer said an algorithm known as K-Means – which is used in a large number of applications on the Internet today – can take large sets of data from large communities and sort out those sub-communities that are more closely related to one another.
For operators using DOCSIS 3.1, this means the chance to assign those user clusters profiles that are a better best fit in terms of bit loading, modulation, error performance or other characteristics.
“Even within optimization, we’ve only really begun to scratch the surface of what these algorithms can do for us in a more general operational context and specifically within the domain of access networks,” Schnitzer said.
Guavus VP of Product and Marketing Chris Menier said machine learning also gives operators the opportunity to distinguish themselves through improved customer experience. And that, he said, is done by better reactions to network issues.
“This is very manual to react to (alerts in the system),” Menier said. “There’s too many of these coming in that you can actually react to, so you have to find some way to take a consistent approach to those alarms, look at them not in a silo but as clusters and then automate some of the actions you can take from them. So we’re moving from this static KPI to really more of a dynamic KPI environment.”
Machine learning can help find impacted populations more quickly, to find commonalities between users and perhaps enable analysis that can take preventative rather than reactive measures to lower the baseline of incidents.
“As you dig under the surface and you look for those micro-populations where you’ve done that identification of those commonalities, you can actually see that you have some micro-populations of customers that are making up that baseline,” Menier said. “When we understand that the baseline is actually made up of a ton of small spikes, of a ton of these paper cuts, of a ton of these micro-populations having issues, that’s when we can really start to bring that baseline down.”