The efficient design of compact antennas operating over multiple bands suitable for the Internet of Things (IoT) is addressed by means of an instance of the system-by-design (SbD) paradigm. More specifically, an iterative strategy that combines different software modules for the search spaceexploration, the fast physicalmodeling of the radiators, and the quality evaluation of the guess solutions is proposed. To enable such a SbD instance, an innovative strategy that exploits an orthogonal array (OA) scheme to determine the training set of a Learning-by-Example (LBE) algorithm based on a support vector regressor (SVR) is introduced for the efficient physicalmodeling of the layout to be optimized. The features and the potentialities of the proposed methodological approach are assessed in different applicative scenarios by considering representative numerical and experimental validation examples.