Unsupervised Learning in Next-Generation Networks: Real-Time Performance Self-Diagnosis

TL;DR

This paper presents unsupervised machine learning methods for real-time performance self-diagnosis in next-generation cellular networks, enabling anomaly detection and relationship learning between performance measures at edge nodes.

Contribution

It introduces two simplified unsupervised learning applications for real-time network diagnosis on edge nodes, focusing on anomaly detection and performance measure correlation.

Findings

Effective anomaly detection and root cause identification

Clustering reveals relationships between performance metrics

Methods operate in near-constant time for real-time use

Abstract

This letter demonstrates the use of unsupervised machine learning to enable performance self-diagnosis of next-generation cellular networks. We propose two simplified applications of unsupervised learning that can enable real-time performance self-diagnosis on edge nodes such as the radio access network intelligent controller (RIC). The first application detects anomalous performance and finds its root cause of faults, configuration, or network procedure failures. The second application uses clustering to learn the relationship between two performance measures. Our proposed applications run in near-constant time complexity, making them, combined with subject-matter expertise validation, suitable real-time RIC applications for network diagnosis.