Due to the high level of magnetic stray field around high power electromagnetic systems, the human exposure needs to be properly assessed in order to check the compliance with international standards and guidelines. Such analyses are usually made in two steps: first a proper map of the magnetic field in the vicinity area is computed, where, in a second time a human model is used to compute induced dosimetric quantities. Unfortunately, such high power systems have a high computational cost in addition to the complexity of 3D human models. Thus, this paper shows the useful combination of stochastic tools with numerical solvers in order to build accurate predictors at a low computation cost in the case of human exposure for various high power systems. These surrogate models can be used to accurately analyze the sensitivity of the exposure problem regarding various input parameters at a low computation cost. A dosimetric methodology for assessing the safety of a human body around an inductive power transfer system for automotive applications, using an adaptive metamodelling algorithm coupled with a voxelized 3D human model, has been developed. Such analysis has been successfully extended to a system where human exposure assessment are crucially needed: medium-frequency direct-current welding guns, treating the case of human exposure to a pulsed magnetic field. This methodology manages to reduce the computation time by more than 99.9% compared to a classical analysis for both exposure problems.INDEX TERMS Human exposure, stochastic methods, numerical dosimetry, metamodel, wireless power transfer, spot welding.