Why your anomaly detection may not be smart enough

This article originally appeared on LinkedIn on February 1st 2021

What is Anomaly Detection?

Anomaly detection (also know as Outlier Detection) is the identification of events or patterns in data which raise suspicions by differing significantly from the majority of the incoming or historical data to which they are compared. We tend to call these events outliers. In the world of telecoms these interesting events are often seen as unexpected bursts in particular types of activity or even a reduction or complete absence of the activity. This presupposes that most data is normal and that burstiness in data can be categorized as outliers worthy of further investigation.

Why Anomaly Detection needs to be smart?

Anomaly detection can be undertaken in many ways, and some are smarter and cater for more compelling use cases than others. For example, imagine you are performing your anomaly detection on aggregated data or data that is not real-time. That data has been collected, processed, aggregated and stored in a DB before you have even started to analyse whether anomalies exist or not. This is certainly slower, more cumbersome and requires that the outliers that are found make sense to act upon sometime after the fact. This lessens the use cases to which this method of anomaly detection can be applied.

Using real-time telecoms data is smarter, for sure, but can you be even smarter again? To make the outlier detection smarter you need to be able to target it to the specifics of the telecom network and you need to know the types of use cases you are looking for in general. Real-time outlier detection can be made inherently smarter by, for example:

• targeting certain algorithms to look for clusters of error-based issues only on the network
• ensuring that other targeted algorithms are looking only for large swings in volume-based metrics on the network, or
• tuning certain algorithms to search only for time-based anomalies on the network.

Why we are better at Anomaly Detection?

Our anomaly detection expertise has been forged in the furnaces of live telecommunication networks with our customers since 2015. We have a patent based on the specifics of the technology. And we have identified use cases common to each of the types of telecom-specific anomaly detection borne out of years of hard work, tuning and expertise.

So what does this mean for Anritsu customers?

Error-based outlier detection algorithms detect anomalous instances concerning network element resourcing issues, congestions and overloads, as well as accessibility, retainability and mobility issues and basically anything that leads to an error being generated on the network.

Volume-based algorithms detect, for example, Wangiri fraud, zero traffic scenarios, signalling storms and more.

And time-based algorithms detect unresponsive nodes, congestion and complex resourcing issues.

Anritsu’s eoMind product delivers smarter anomaly detection, finding both the common and uncommon issues on all layers of the telecoms networks.