EdgeMCS Advisor: Real-time MCS Selection and Bitrate Guidance at the MEC Edge

This paper presents EdgeMCS Advisor , an edge-deployable pipeline that converts radio key performance indicators (KPIs) observed in Licensed-Assisted Access (LAA) cellular networks into real-time guidance for downlink modulation and coding scheme (MCS) selection and application-layer bitrate control at the mobile edge computing (MEC) layer. Using a public IEEE DataPort LAA measurement dataset, we train supervised multi-class models to predict the codeword 0 MCS family (QPSK, 16QAM, 64QAM, or 256QAM) from SINR, CQI, BLER, resource-block allocation, and throughput counters. After rigorous cleaning, feature engineering, and stratified cross-validation, a histogram-based gradient boosting classifier achieves \((0.994)\) accuracy and \((0.962)\) macro-F1 on a held-out test set, and its probability outputs are assessed with calibration curves. We package the trained model and preprocessing steps as a FastAPI microservice with a lightweight FastHTML interface and a SQLite-backed decision store that logs predictions, calibrated per-class probabilities, and optional MEC-host telemetry. A recommendation endpoint combines model confidence with explicit QoS descriptors (traffic class, latency budget, and bitrate bounds) to return an MCS suggestion, bitrate headroom, a packet-size heuristic, and a qualitative risk indicator. The prototype demonstrates a practical path from RF telemetry to QoS-aware control loops that can run at the MEC edge in latency-sensitive deployments such as smart factories and stadiums.