What is an anomaly?
Essentially, something out of the ordinary, unexpected, or radically different from what has been predicted can be said to be an anomaly. One might think that if this is the case, it should not be hard to find. At the end of the day, it should stick out like a sore thumb. However, given the huge amount of data that passes through networks, it is more like finding a needle in a haystack! Not only this, but as there are often multiple anomalies, you need to be able to detect those that have more impact on the customers and are of top priority.
Fortunately, given the advancement in technology people have been able to build machines and automated software, combined with very powerful and low-cost computing, that can handle data sets at speeds that people simply cannot. How can Artificial Intelligence (AI) and Machine learning (ML) help operators catch anomalous behavior?
Types of anomaly detection:
There are various methodologies for how AI/ML can be used to detect outliers, including supervised, semi-supervised, and non-supervised methods.
Supervised detection is by nature the most labor-intensive, as you would need a person to be able to establish within a data set 2 categories – normal and abnormal. As the machine is identifying trends from human assignment, one needs to make sure that the training data sets are of high quality so that learning mistakes do not creep in. If this occurs, this method becomes very reliable for catching outliers.
Unsupervised detection requires no manual laboring when it comes to labeling data sets. The benefit is of course that this is very time efficient, however, there is a downside. Creating such a system is very complex. Since the learning is unsupervised, there is a greater probability that the AI could make learning errors. Due to these considerations, unsupervised detection is not as reliable as supervised detection.
Semi-supervised detection is a mix of the 2 methods above. Especially when dealing with a large amount of data, one needs to find the right balance of accuracy on the one hand and efficiency on the other. Controlling how AI learns, while at the same time applying unsupervised learning methods to automate feature learning with unstructured data, often allows for the best of both worlds.
Aspiring for perfection:
AI can significantly reduce the time engineers spend looking for issues and allow them to spend more time optimizing the network. Today’s detection tools need to be highly reliable in order to give confidence to operators and their customers. False reports can have a really serious impact, which is why assurance capabilities have never been more important. Using a mix of supervised and unsupervised detection methods helps spot and deal with anomalies.
There is a serious business case for the increased use of AI and ML. A white paper1 published by Mobile World Live has shown that “more than two thirds of global operators recently surveyed by GSMA Intelligence said they currently sell private wireless networks specifically deployed for enterprise customers. The vast majority of these operators have at least ten enterprise customers; 41 per cent have at least 50.”
It may not always be easy for operators to implement such technologies, but the world is heading into the world of automation. Since AI is the building block for automation, it comes as no surprise that as the use of AI and ML increases, so will the use of automation in numerous capacities. Despite this, in certain segments, there is confusion around this technology. How can one really know what is going on inside of the networks and data streams? Can any meaningful, actionable insights be gleaned from such huge amounts of information?
The answer to these questions is yes! We can find out what is happening inside all that traffic, if there are any meaningful patterns, as well as odd results and converting insights into real time action. To this end, the world of assurance has never been more important. As a key player in the 5G assurance market, RADCOM has created RADCOM ACE and RADCOM AIM that are designed for just this purpose. The RADCOM AI Module, (RADCOM AIM), is an AI capability that can be added to RADCOM ACE to engage in anomaly detection, or it can be the foundation for RADCOM NWDAF.
Automated assurance, containerized, and end-to-end analytics allow you to see exactly what is going on in your network. You can use RADCOM AIM as a foundation for the RADCOM Network Data Analytics Function (NWDAF) and transition to a closed-loop approach to network management. This will allow you to analyze and implement strategies to improve your network’s performance, stability, function, and ultimately your customer experience.