RAN Explainable Anomaly Prediction for 6G Networks

Machine Learning (ML) models integrated into future 6G network architectures are anticipated to address the demanding task of Anomaly Detection (AD) at the Radio Access Network (RAN). A challenging but essential aspect of this task is forecasting anomalies ahead of time, aiming to enable automated interventions to proactively prevent or mitigate service degradation. To address this challenge, we propose an Explainable Anomaly Prediction (XAP) framework, that combines a time series forecasting Long Short-Term Memory Encoder- Decoder (LSTM-ED) model, and an Isolation Forest (IF) anomaly detector whose output is interpreted using the SHAP algorithm. The framework is evaluated using a real- world dataset containing RAN Key Performance Indicators (KPIs) collected from a User Equipment (UE) device. The results validate the effectiveness of our framework in predicting RAN anomalies and identifying the specific KPIs linked to each predicted anomaly.