An End-to-End Smart Radio Log Anomaly Detection Architecture for 5G/6G Networks

In the era of 5G/6G networking, the complexity and scale of modern cellular networks have increased significantly. Consequently, the volume of generated logs has also multiplied, making comprehensive anomaly detection a cumbersome task for operators. In response, this article presents the Radio Log Anomaly Detection (RLAD) architecture, an MLOps-driven pipeline designed to enable automated, end-to-end anomaly detection in log data from next-generation networks. The key contribution lies in a design that integrates continuous training, deployment, and performance monitoring mechanisms to mitigate data drift and ensure model relevance in dynamic telecom environments. The architecture also addresses the practical challenges of labeled data scarcity and class imbalance, enabling robust detection even in unsupervised settings. As a proof-of-concept, we instantiate the pipeline with a lightweight LSTM autoencoder enhanced with attention, achieving 98% accuracy, 99% F1-score, and a 99% precision while outperforming a baseline LSTM. The system delivers anomaly insights via a user-friendly interface that supports operator diagnostics and feedback integration.