AI and ML to detect anomalies, enhances QoE and bridge the skills gap

The combination of AI and ML-based anomaly detection techniques with recent technology breakthroughs like digital twins and Large Language Models (LLM) can solve complex real-world problems in the telecoms industry claimed Benoît Drooghaag, AI/ML Application Expert, Nokia. Access to data is key to unlock the power of AI and ML. Modern telemetry continuously monitors and measures real-time network parameters such as alarms, traffic utilization or equipment metrics. The process can be used to extract key patterns and trends in data, automatically detect anomalies in the network and suggest potential solutions using proactive remediation.

To determine anomalies in the network, AI and ML algorithms can be retrained periodically on the network’s previous data sets and they can be tailored to the service provider’s own metrics and KPIs.

Drooghaag expressed that anomaly detection is nothing without closed-loop automation and root cause analysis as these are the processes that provide potential solutions. “Detecting anomalies is just the first step, it is vital to identify their root cause and take appropriate action to solve them, especially if it impacts the customer’s Quality of Experience (QoE).”

For example, there could be different underlying reasons for the cause of congestion in a network and a number of varying solutions could be identified advised Drooghaag. “If a plateau is highlighted in a subscriber traffic, suggesting a service tier limit has been reached, this could present operators with the opportunity to upsell more services.” This is important, not only for reducing customer churn, but for the opportunity to increase average revenue per customer (ARPU).

In a near future, the introduction of technologies such as digital twins and LLM-based AI agents inside the control loop will enable the network controller to make autonomous decisions to achieve the most optimal results. For example, it will be possible to automatically identify the most effective configuration changes to be applied to the network to enhance data traffic. This approach is beneficial in complex scenarios where the number of potential configuration changes is vast and dynamic.

Ultimately, anomaly detection – enabled by AI and ML – collects and analyzes network data to proactively detect any potential issues before they become real problems. Using a physical copy or simulator of the real-life network, known as a digital twin, can provide an in-depth and unbiased view of the entire network and its performance. It also ensures that less risks are taken concerning network configurations. The network operator can input potential solutions suggested by AI into the digital twin to test and evaluate the network behaviour before applying them to the real network.

A digital twin is not just a simulation tool, said Drooghaag. “A digital twin is a complete platform providing network design services, not only for network operators, but on field engineers for all deployment processes.”

A combination of AI/ML-based anomaly detection, digital twins and AI-agents trained on domain expertise can help reducing the skill gap in the telecom industry by increasing the automation and abstraction level, leading to faster and easier problem resolutions, increased network deployment speeds, and an enhanced customer QoE and higher operator ARPU.