CQI Prediction in 5G Networks Using a Mobility-Aware Random Forest Model

1. Introducing a random forest-based machine learning model for CQI estimation that incorporates the velocity of the user along with traditional radio parameters SNR, RSRP, and RSRQ.
2. An extensive comparison of the proposed model against 9 benchmark algorithms, demonstrating superior performance in metrics like RMSE and R², validating its effectiveness over state-of-the-art methods.
3. A comprehensive ablation study that shows the impact of including velocity in CQI estimation, showing an almost 13% decrease in RMSE, statistically validating its importance.
4. The model’s robustness has been validated by showing consistent performance across both low- and high-velocity scenarios.
5. Local Interpretable Model-Agnostic Explanations (LIME) has been employed to explain the decision-making process.
6. Beyond error metrics, the real-world impact of the proposed model has been assessed using feedback overhead analysis and noise robustness, showing the RF-predicted CQI yields better fairness and a robust outcome compared to traditional SINR-to-CQI mapping.