A Machine Learning Framework for Enhancing <scp>5G mmWave</scp> Radio Frequency Prediction

ABSTRACT 5G mmWave technology offers high data rates and bandwidth, but high path loss and environmental changes affect signal quality. Existing models are not suitable for mmWave channels due to their varying nature over time. To overcome these challenges, this research presents an efficient time‐dependent channel modeling framework based on a cuttlefish search‐inspired efficient support vector machine (CS‐ESVM) for predicting channel characteristics in large‐scale measurement and RF at specific measurements at LOS and NLOS. The model is proposed to work for measurements at 28 GHz at a substation. The model also combines a prediction model and playback model for accurate channel characteristics key metrics such as root mean square error (RMSE), mean absolute percent error (MAPE), and correlation coefficient (CC), predicting the radio frequency. The proposed CS‐ESVM model achieved the lowest RMSE values of 2.510 (LOS corridor), 1.210 (LOS hall), 1.815 (NLOS corridor), and 1.917 (NLOS hall), the lowest MAPE values of 0.009, 0.004, 0.003, and 0.007, and the highest CC values of 0.899, 0.969, 0.921, and 0.985. The findings suggest that CS‐ESVM is more effective at predicting the mmWave channel's characteristics than traditional approaches. In conclusion, this ML‐based framework improves the projection of 5G mmWave RF channels and provides a stable solution for real‐time prediction in future network environments.