Customers notice service quality issues within minutes. What’s more, according to a PwC study, 32% of customers would stop doing business with a brand they love after one bad experience. That really puts a premium on quality.

Broadband is no exception. Customers expect broadband technology to always work and don’t take any notice of it - unless it malfunctions. This means network performance, quality and availability are key. In another blog post we presented an Altiplano app that improves the availability of the subscriber peak speed to improve the subscriber experience. Here, I’d like to showcase the Network Trend Analyzer (NTA), an innovative app which makes it easier to detect network events that do not correspond to a normal, healthy behavior.

To avoid always resolving issues in a reactive manner, you want to identify as much as possible network problems before the customer reports them. Otherwise, you face the urgency of truck rolls, scarcity of spares and customer discontent. But proactivity comes with a challenge: how to accurately identify patterns that forewarn network issues?

Traditionally you would use rule-based analytics like TCA (Threshold Crossing Alerts) where the system can raise notifications when the KPI values cross a set threshold. However, thresholds are usually set manually, defined network-wide and designed to deal only with extreme conditions. What’s more, the TCA rule needs continuous manual tuning to keep the thresholds accurate. If you make thresholds too conservative, you'll have a lot of false negatives (missed detections). Setting the threshold too aggressively will lead to an increase of false positives (wrong detections) that will make the system unreliable. That's why you need something better.

The NTA uses machine learning techniques to derive the trend of monitored KPI – without relying on any presets - using its historical characteristics including fundamental trend, random variance, seasonal, weekday and time of day behavior.

This trend helps detect real-time anomalies by identifying short term deviations with respect to the expected trend. This technique is more reliable and shows high accuracy. Network events could include system temperature, optical attenuation, traffic spikes, or unusual CPU/memory/resource usage.

NTA also allows you to do medium-term threat prediction and predict future issues. Thus, anomalies can be reported before any actual occurrence of the issue. Examples could be a gradual increase in board temperature which could become problematic in a month’s time (see the video illustration), UPS battery cells degrading, or increasing PON occupancy that could result in contention. This allows operations teams to plan mitigation, schedule field visits and fix issues before they affect the all-important customer experience.
Thanks to machine learning capabilities, NTA trains the trend continuously, thus capturing nuances and continuously improving prediction accuracy. We illustrate this in a video which shows the operational board temperature. The algorithm needs a few weeks of data to correctly capture the trend, but once the history of the metric is built up, the NTA is able to predict already in August that the operational temperature threshold of 75 degrees Celsius will be crossed in September - if no preventive measures are taken.

Video: The anomaly detection algorithm of the Network Trend Analyzer at work

NTA offers flexibility for operators to monitor one or more KPIs of their choice and provides a visual indicator and dashboard of predicted vs actual trend. If you’d like a preview or learn more about NTA, visit the Altiplano Application Marketplace.