Machine Learning-Based Anomaly Detection and Root Cause Analysis in Mobile Networks

This paper proposes an automatic anomaly detection and root cause analysis model that combines Density-based Spatial Clustering of Applications with Noise (DBSCAN) and Long Short-Term Memory (LSTM) Auto-Encoder (AE). The model is trained and tested using unlabelled real-world 4G network data from a Portuguese operator. The study demonstrates the promise of DBSCAN in separating normal network traffic patterns from abnormal ones, as well as the ability of LSTM-AE to learn daily network Key Performance Indicator (KPI) behaviour and detect anomalies and their probable root causes based on reconstruction errors. Results indicate that anomalies in available network KPIs do not always result in abnormal traffic patterns, and vice versa. Consequently, it can be inferred that relying solely on traffic volumes for DBSCAN is not an ideal method to separate normal network data from abnormal data for detecting network anomalies. Additionally, the results emphasize the importance of high-quality data in terms of sampling rate and the number of KPIs, as well as the significance of data analysis in identifying patterns across different levels of mobile networks.