Machine learning prediction model for the sewing thread consumption

This study aims to provide rapid and precise methods for industrial users to predict the amount of sewing thread required to sew garments using different stitches of class 600. To avoid unused stocks, sewing consumption for each stitch was computed using the Extra Trees machine learning method. Multiple machine learning models including Extra Trees, Support Vector Machine, CatBoost, Random Forest, Linear Regression, and Artificial Neural Networks were compared. The Extra Trees model achieved superior performance with mean absolute error (MAE) of 4.24 cm, root mean squared error (RMSE) of 5.05 cm, mean absolute percentage error (MAPE) of 1.81%, and R² of 0.974. In accordance with the findings, the prediction of sewing thread consumption related to each stitch was proposed as a function of the studied input parameters (Plies and Stitch Length). Extensive cross-validation (5-fold cross-validation) showed that the performance is strong and the overfitting is low. Extra Trees was the best model among all others, including Support Vector Machine (R² = 0.939), CatBoost (R² = 0.898), and Random Forest (R² = 0.884). These findings support the practical use of the Extra Trees ensemble approach for industrial thread consumption prediction, and the stringent validation protocol acts as a guarantee that the approach will be applicable to novel data.