Comparative machine learning for multi-response quality prediction in CO ₂ laser cutting of polypropylene

Abstract This research study presents a comprehensive comparative analysis using machine learning techniques to predict multiple quality characteristics in CO 2 laser cutting of polypropylene. A full factorial experimental design was employed with three process parameters: focal length (6.5–8.5 mm), laser power (85–100 W), and cutting speed (4–12 mm s −1 ). The effects of these parameters on surface roughness Ra, top and bottom kerf widths KW, and kerf angle KA were systematically investigated. Experimental results show that cutting speed has the strongest influence on Ra, while kerf geometry is mainly governed by focal length. Moreover, KA is significantly affected by the interaction between cutting speed and focal length. The machine learning models that were created using standardized data sets are Extreme Gradient Boosting (XGBoost), Categorical Boosting (CatBoost), Gradient Boosting Regressor (GBR), H2O Gradient Boosting Machine (H2O_GBM), and Gaussian Process Regression (GPR). Model performance was evaluated using the coefficient of determination, root mean squared error, mean absolute error, and Pearson correlation coefficient. Ensemble-based models, particularly XGBoost and CatBoost, achieved the highest prediction accuracy for most responses, whereas GPR performed best for Ra. The results indicate optimal cutting quality is achieved at high cutting speeds, large focal lengths, and elevated laser power under controlled processing conditions.