Rakuten Symphony has developed a prioritized list of the top 25 AI initiatives for communications service providers and divided them into five categories. The Network Performance Anomaly Detection use case described in this blog falls under level 3 automation of anomaly detection.
The primary goal of Rakuten Symphony’s Network Performance Anomaly Detection use case is to use AI/ML to automate the monitoring of network performance KPIs to identify unusual pattern in the performance (anomalies). MNOs can use these as detectors to identify and catch problems early and help MNOs improve their quality of services.
For each important system in the network, network engineers establish a number of key performance indicators (KPIs) to help track performance of that system. If the KPIs are not being met, the engineer knows there is a network problem that must be reported to the right engineering team for fixing.
Many MNOs still do this manually. Some of the issues with this approach include:
The Network Performance Anomaly Detection use case is designed to use AI / ML to track KPI data in real time, forecast the performance status of the KPIs, and assign an anomaly score for any KPI that is not performing. This score indicates the severity of the anomaly. This process allows delegation to the responsible repair team and assigning the proper repair priority level. The speed of the AI process enables more issues to be corrected before a system fails.
This use case is part of the Rakuten Symphony Performance Monitor software that tracks network health and performance by collecting raw data from network devices or environment monitoring systems (EMS). It is highly customizable through user-defined KPIs and thresholds and a business intelligence framework for reporting functions and dashboards.
For the AI training and inference, the Network Performance Anomaly Detection use case connects via an API to the Rakuten AI platform, which provides intuitive tools for experts across domains and teams to collaborate around, centralize, and streamline data management and AI modeling processes. It supports both predictive AI and generative AI models.
The AI models for this use case are capable of:
The Network Performance Anomaly Detection use case captures time series patterns across domains and geographies and can monitor a large number of KPIs that enables the model not only to detect the anomalies within single KPI but also give insight on potential KPIs that are contributing to those anomalies.
The use case also features a user friendly dashboard that empowers operators to quickly and proactively respond to failures.
Network Performance Anomaly Detection has been deployed in the Rakuten Mobile network for more than a year and has proven its value with a growing number of deployments with different teams within the MNO.
The use of KPIs to understand network performance is an effective way to determine the operating status of a network and any conditions that might make it fail. But manual processing of this data is slow and error prone. Every moment the network is impacted can result in customer dissatisfaction that results in churn. With the Network Performance Anomaly Detection use case, Rakuten Symphony has leveraged the power of AI/ML to understand the performance issues – accounting for seasonality – and alert the correct team to resolve the underlying problem. The impact of this use case can be a significant improvement in mean-time-to-repair, leading to more satisfied customers.
