Deep Learning-Based Log Anomaly Detection for 5G Core Network

Abstract

Adopting network function virtualization in 5G core network (CN) enables flexible and agile service development, but it also brings increased complexity and the likelihood of anomalies, emphasizing the vital importance of effective anomaly detection. While existing research primarily focuses on detecting external anomalies for 5G networks through network traffic analysis, there is a growing need to identify internal abnormalities and failures within the 5G CN. Towards this end, considering the wide recognition of log data as a valuable information source for troubleshooting and fault diagnosis, we develop a deep learning (DL)-based log anomaly detection framework for 5G CN. The framework encompasses log parsing, log grouping, feature extraction, and model training, and each module is designed with distinct functionalities to enable combinational usage in various situations. We also establish a cloud-native 5G testbed to facilitate the collection of a large-volume 5G CN log dataset, wherein multiple types of anomalies are injected. Evaluation results illustrate that our highest achieved F1 score exceeds 97%, highlighting the effectiveness of our proposed anomaly detection framework for 5G CN.