Lab: Design an AI/ML pipeline for anomaly detection in RAN KPIs

Pipeline Design: Data Collection and Preprocessing

This lab walks through designing an end-to-end AI/ML pipeline for detecting anomalies in RAN KPIs. The pipeline begins with data collection from the gNB via PM counters (3GPP TS 28.552) at 15-minute granularity: RSRP distribution, DL/UL throughput, PRB utilization, handover success rate, RRC setup success, and RACH failure count. Preprocessing steps include handling missing data from element outages (interpolation vs exclusion), normalizing counters across different cell configurations (macro vs small cell), and engineering derived features such as busy-hour ratios, week-over-week deltas, and…