Copyright © 2025, IEEECublier Martínez, AymarFrövel de la Torre, JorgeSanz, RuyPlaza Gallardo, BorjaPoyatos Martinez, David2026-03-172026-03-172025-112025 Antenna Measurement Techniques Association Symposium (AMTA) - Tucson, AZ, USA: 1-6978-1-7362351-7-12474-2740https://hdl.handle.net/20.500.12666/1777IEEE Keywords: Machine learning algorithms, Temperature, Permittivity measurement, Training data, Measurement techniques, Predictive models, Prediction algorithms, Data models, Electromagnetics, PermittivityThe introduction of composite materials in the aerospace industry has yielded improved performance and better strength-to-weight ratio in space platforms' design. The ubiquity of these materials in today's space missions involves, among other things, a deep understanding of their electromagnetic (EM) properties. To study the EM properties of dielectric materials, several machine learning algorithms such as Deep Neural Networks (DNN) and ensemble learning methods (Random Forest, Gradient Boosting) are implemented as extraction methods for EM characterization in free space. First, the training and testing datasets for the supervised learning algorithms are generated using EM simulations with given permittivity and loss tangent. Second, an existing high precision test bench is used at INTA's Computational and Applied Electromagnetics Laboratory (CAEM-Lab) for gathering data to feed the prediction models at the validation stage over a broad frequency range of 2.6−40GHz. Finally, results are contrasted with conventional methods for complex permittivity extraction, such as analytical based solutions and optimization related methods. Other features like temperature can be added to the models to study the EM properties of materials at cryogenic temperatures for space applications.engcomplex permittivitydielectricelectromagnetic characterizationloss tangentmachine learningmaterialsspace applicationsinfo:eu-repo/semantics/conferenceObject10.23919/AMTA66658.2025.11317850info:eu-repo/semantics/restrictedAccess