Performance evaluation of regression models for predicting dimensional stability of heat-treated silver fir wood based on mass loss, contact angle, and color changes

This study aimed to predict water absorption (WA) and swelling of heat-treated silver fir wood (Abies alba) at 180, 200, and 220 oC by simple regression, multiple linear regression, as well as multiple non-linear regression models. ∆E (total color difference), ∆L (lightness difference), contact angle (CA), and mass loss (ML) were used as predictors. The results showed that WA, volumetric swelling (SV), swelling in longitudinal, radial, and tangential directions (SL, SR, and ST) decreased with the increase of heat-treatment temperature, but the values of ∆E, ∆L, CA, and ML increased. The lowest mean absolute percentage error (MAPE) for the prediction of WA with the simple regression models was related to the Cubic model based on ML equal to 6.22. The lowest MAPE for the prediction of SV, SR, and ST were related to the Cubic model based on ML equal to 3.01, 3.55, and 4.2, respectively. The MAPE values for the prediction of WA, ST, SR, and SV by the multiple linear regression model were 6.11, 3.9, 3.89, and 2.7, respectively. Their corresponding values for the non-linear regression model were 5.76, 3.86, 3.6, and 2.61, respectively. Since MAPE below 10% is satisfactory for predicting, the studied models have predicted WA and their corresponding swelling of heat-treated wood with acceptable accuracy. The best time and cost-efficient regression model is the simple model, and the best predictor in terms of time, cost, and the ability to measure in line is the color index.