Prediction of the flexural strength of particleboard using artificial neural network modeling in comparison with regression models

Today, several modeling methods have been used to cost-efficiently predict the physical and mechanical properties of wood-based panel products which in turn reduce the cost of quality control of these products. Two common methods include regression and artificial neural network (ANN). In this study, the possibility of predicting the modulus of rupture (MOR) and modulus of elasticity (MOE) of particleboard by simple and multiple linear regression and ANN based on structural parameters including density in three levels (0.65, 0.7, and 0.75 g/cm3), slenderness ratio of particles in three levels (47, 30, and 13) adhesive percent in three level of (8, 9.5, and 11%) were evaluated. experimental and predicted data were compared with different criteria including mean absolute percentage error (MAPE), mean squared error (MSE) and coefficient of determination (R2). The results revealed that although both multiple linear regression models and artificial neural network were able to predict MOR and MOE values with acceptable accuracy, but ANN model predicted them with higher R2 and lower MAPE than multiple linear regression model. The value of MAPE and R2, for prediction of MOR and MOE by ANN model were 7.72% and 0.77, and 7% and 0.86, respectively. the corresponding value for multiple regression model were 8.3% and 0.738, and 9.06% and 0.783, respectively. These levels of error are industrially and practically satisfactory for the prediction of properties in particleboard.