Building an anomaly detection xApp: KPM → ML inference → alerts

Anomaly Detection Pipeline Architecture

An anomaly detection xApp applies machine learning to real-time RAN metrics to identify abnormal conditions before they impact subscribers. The pipeline begins with KPM subscriptions collecting per-cell and per-UE measurements — RSRP, SINR, throughput, retransmission rate, and timing advance. These raw metrics flow into a feature extraction stage that computes rolling statistics: mean, variance, percentiles, and rate-of-change over sliding windows stored in SDL. The engineered features feed an inference engine running a pre-trained model that classifies each measurement window as normal or…