In this study, the CPA algorithm was used to optimize a BP neural network model to predict the bond strength and surface roughness of heat-treated wood. The neural network model was trained and optimized using MATLAB software. The results of the BP neural network, random forest algorithm, and optimized CPA-BP model were compared. The results show that the CPA-optimized BP neural network model has a better R² compared to the conventional BP neural network model. After using the CPA-optimized BP neural network model, the R² value increased by 8.1%, the MAPE value decreased by 3.74%, and the MAE value decreased by 33.91% in the prediction of the surface bond strength. The R² values increased by 3.02% and 20.47%, respectively, in predicting the mean and maximum values of surface roughness. The results indicate that the model is reliable in predicting wood bond strength and wood surface roughness. Using this model to predict wood bond strength and surface roughness can also reduce the required experimental cost.