Interpretable machine learning for laser machining morphology prediction of CFRP driven by physical-derived features

To address the limitations of CFRP laser machining process prediction in methodological benchmarking and mechanistic interpretability, this study proposes a morphology prediction framework that simultaneously integrates point-prediction accuracy, uncertainty quantification, and interpretability. Physically derived features are introduced to bridge external process parameters and morphological responses through a causally constrained energy-flow pathway, while a concise and reliable model is identified through systematic evaluation. Six baseline machine learning models are comparatively assessed along two principal dimensions—accuracy and uncertainty. By incorporating the coefficient of variation, maximal information coefficient, and recursive feature elimination, physical features exhibiting low dispersion, low collinearity, and high importance are selected to construct a three-layer causal chain of raw process parameters–physically derived features–morphological indicators. A dual-layer SHAP analysis is subsequently employed to hierarchically delineate the contribution pathways from process parameters to morphological responses.The results demonstrate that Gaussian Process Regression outperforms the other models in both predictive accuracy and uncertainty representation. Compared with models using only raw features, the inclusion of physically derived features enhances the reliability of uncertainty characterization and establishes physically constrained causal linkages between process parameters and morphological indicators. The three-layer causal chain, combined with the dual-layer SHAP analysis, jointly elucidates the distributional patterns and mechanistic contributions of morphological responses, thereby strengthening the causal consistency and interpretability of the predictive model. This work provides an efficient, robust, and interpretable technical paradigm for morphology prediction and process optimization in CFRP laser machining.