Akademik Veri Yönetim Sistemi
Measurement: Journal of the International Measurement Confederation, cilt.272, 2026 (SCI-Expanded, Scopus)
The accurate determination of complex dielectric properties of natural materials, particularly wood, is essential for applications in industrial heating, sensing, and sustainable electronic systems. In this study, both measurement-based and data-driven approaches were combined to investigate the real (ε′) and imaginary (ε″) components of complex permittivity (ε*) for six hardwood species named walnut, sycamore, aspen, chestnut, oak and eucalyptus commonly used in Turkey. Microwave measurements were conducted in the S and C bands under varying frequency, density, and moisture conditions to construct an original dataset collected over the course of one year. A microwave waveguide-based measurement approach was employed, in which s-parameters were recorded over the 2–7.1 GHz frequency range and used to compute the frequency-dependent electrical properties of samples. The resulting experimental dataset comprises 99,000 s-parameter measurements, which were used to extract four distinct parameters (ε′, ε″, ε*, LT) across three independent waveguides, each sampled at 500 frequency points. This comprehensive dataset enabled the development and evaluation of eight machine learning regression models for predicting dielectric properties. Among the evaluated models, the Extra Trees achieved the highest predictive accuracy on the test set for predicting ε*, with a mean squared error (MSE) of 0.98 and a coefficient of determination (R2) of 0.70. Furthermore, cross-validation analysis confirmed the robustness of the model, yielding an average MSE of 0.94 across the validation folds. On an independent verification dataset, the proposed model maintained strong generalization performance, achieving an MSE of 3.99 and an R2 score of 0.64 for ε* prediction. The results confirm the consistency of the microwave waveguide measurements and demonstrate that data-driven machine learning models, particularly the Extra Trees Regressor, provide improved robustness and predictive accuracy compared with classical regression approaches. Overall, this study demonstrates that frequency and density measurements are sufficient to estimate the complex dielectric permittivity of hardwood specimens accurately, providing a reliable and non-destructive framework that bridges experimental microwave characterization with modern machine learning techniques.