The deployment of near future 5G networks will introduce modifications in the population’s exposure levels to radio-frequency electromagnetic fields (RF-EMFs). The present work aimed to face the challenge of studying the exposure variability in the presence of an access point (AP) at 3.7 GHz with 64 patch elements uniform planar array antenna and 3D beamforming capability. The novelty introduced in the methodology of the exposure’s evaluation was the combining of traditional computational methods with a new approach based on stochastic dosimetry, called polynomial chaos kriging method, in order to estimate the exposure levels for 1000 different antenna beamforming patterns with low computational efforts. The simulations were evaluated considering a child model and computing the specific absorption rate (SAR) in different tissues. The analysis of the results highlighted a high exposure variability scenario depending on the beamforming patterns of the array antenna and identified the ranges of elevation and azimuth angles of the main antenna beam that may cause the highest levels of exposure.