Interpretative analyses for milling surface roughness prediction in thermally modified timber: Shapley value (SHAP) and local interpretable model-agnostic explanations (LIME)

As a green building material, thermally modified timber's milling surface roughness significantly impacts processing quality, yet the influencing process parameters are complex and interdependent. This study predicted milling surface roughness using modification temperature, depth of cut, feed rate, and spindle speed as input features. Among the compared models, XGBoost outperformed Random Forest in prediction accuracy. The study further integrated Shapley value (SHAP) and local interpretable model-agnostic explanations (LIME) to analyse the model's interpretability. Results revealed that the depth of cut had the most significant impact on surface roughness, followed by feed rate, modification temperature, and spindle speed. Local interpretability analyses provided detailed insights into each parameter's contribution to single-sample predictions. This research demonstrates the value of SHAP and LIME in manufacturing, enabling transparency in machine learning models and offering theoretical support for process control in automated systems. The findings provide practical guidance for optimising timber processing parameters and advancing real-time monitoring and optimisation in smart manufacturing.