Comparison of Anomaly Detection Methods and Root Cause Analysis in Radio Access Networks

Radio access networks (RAN) produce large amounts of operational data, making manual performance monitoring difficult and inefficient. This study compares three automated anomaly detection methods: statistical methods, deep learning models, and rule-based approach. Statistical methods work well for detecting short-term variations, but they often give false alarms. Deep learning models excel at identifying complex or long-term collective anomalies, but they require substantial computing power and extensive training data. A rule-based approach accurately detects known and clearly defined issues but struggles with new or evolving cases. Although these methods can identify abnormal behaviour, finding the exact cause is still challenging. Anomalies may be due to planned maintenance, hardware failures, or incorrect configurations, and determining the cause typically requires additional information and expert knowledge. The results suggest that combining these methods is most effective, as each technique detects different types of anomalies and provides better overall coverage.