Radio-frequency electromagnetic field (RF-EMF) exposure characterization involves the determination of the levels of RF electromagnetic fields incident on humans in lab or real environments. The most important sources are antennas for broadcast and wireless telecommunication networks. Understanding the exposure of individual people to such fields over time is of crucial importance for studying health effects and checking compliance with norm and regulations. The classic exposure characterization methods are based on a massive amount of measurements in the streets which are very time consuming and expensive. To make it practical, the number of measurements is often limited to a small random set that can be executed in reasonable time, although this means that important variations in the field could be missed due to undersampling. One tries to improve the results by using prior knowledge of the environment;however  this information is often inaccurate, unavailable or simply does not match with the reality.

Therefore, reliable characterization of RF-EMF exposure is still an issue to be investigated.

To tackle this problem, a new methodology will be investigated that improves the reliability and cost-effectiveness of electromagnetic exposure characterization in a complex environment. This methodology will be based on an effective interplay of machine learning, surrogate modeling techniques, and optimization. So far, this approach is very innovative in the context of RF-EMF.