We propose a methodology for developing EM-based polynomial surrogate models exploiting the multinomial theorem. Our methodology is compared against four EM surrogate modeling techniques: response surface modeling, support vector machines, generalized regression neural networks, and Kriging. Results show that the proposed polynomial surrogate modeling approach has the best performance among these techniques when using a very small amount of learning base points. The proposed methodology is illustrated by developing a surrogate model for a T-slot PIFA antenna simulated on a commercially available 3D FEM simulator.
