Machine learning-based prediction of some mechanical properties of sycamore maple (Acer platanoides L.) wood at varying moisture levels

This study systematically investigates the mechanical response of Norway maple (Acer platanoides L.) wood conditioned at equilibrium moisture contents of 0%, 12%, 27%, and 50%. Modulus of rupture (MOR) and modulus of elasticity (MOE) tests, as well as compressive strength parallel to the grain (CS//) tests, were conducted on the prepared samples. Subsequently, machine learning techniques such as Random Forest (RF) regression, Polynomial Regression (PR), and Support Vector Regression (SVR) were used to predict the mechanical response of the wood. As moisture content increased from 0% to 50%, bending strength decreased by 65.15%, and LPBD decreased by 60%. These results clearly demonstrate that moisture content has a significant effect on the mechanical strength of white ash wood. The RF model demonstrated strong generalization performance, maintaining a consistently high coefficient of determination (R² ≈ 0.90) across all moisture conditions. PR provided significantly accurate deflection predictions (R² between 0.780 and 0.999) at various moisture content levels and notable variations in elasticity predictions. The SVR model provided significantly accurate predictions for deviation (R² = 0.806–0.986), but notable differences were observed in elasticity predictions (R² = 0.585–0.971), particularly in regions close to the fiber saturation point.