Machine Learning Modeling for Radiofrequency Electromagnetic Fields (RF-EMF) Signals From mmWave 5G Signals

5G is the next-generation mobile communication technology that is expected to deliver better data rates than Long-Term Evolution (LTE). It offers ultra-low latency and ultra-high dependability, enabling revolutionary services across sectors. However, 5G mmWave base stations may emit harmful radiofrequency electromagnetic fields (RF-EMF), raising questions about health and safety. Our research suggests that the RF-EMF prediction model lacks sufficient papers or publications. Therefore, this study employs IEEE and ICNIRP standards for assessment and exposure limits. The measuring campaign analyses one sector of a 5G base station (5G-BS) operating on 29.5 GHz in Cyberjaya, Malaysia. This study proposes two prediction models. The first model predicts the signal beam RF-EMF, while the second predicts the base station RF-EMF. Each model contains three machine learning techniques to forecast RF-EMF values: Approximate-RBFNN, Exact-RBFNN, and GRNN. The results are analysed and compared with the measured data, determining which algorithm is more accurate by calculating the RMSE of each algorithm. As a result, it can be observed that the Exact-RBFNN algorithm is the best algorithm to predict the RF-EMF because it shows good agreement with the measured value. Moreover, in a 1-minute duration, the difference between the predicted and measured values reached 0.2 less channels. However, in 6 minutes and 30 minutes, it can observe more accurate results since the differences between values reach 0.1 in these situations. Additionally, the ICNIRP standard was used and compared with the results and validation values of the algorithms.