Machine Learning-Based Prediction of Yarn Breakage in the Sizing Process of Cotton Fabrics

Abstract Yarn breakage during the sizing stage of cotton fabric production process leads to yarn waste, increased loom downtime, and energy losses, thereby reducing profit margins. This study uses machine learning methods to predict the yarn breakage rate per kilometer during the sizing process in order to make decisions aimed at reducing waste and cost. A comprehensive parameter set was created by combining yarn mechanical properties such as yarn tenacity and elongation at break with machine speed, exit moisture, drum temperature, and various stress values. Several advanced supervised learning algorithms, including decision trees, neighborhood-based, kernel-based, and ensemble tree methods, were compared. Seven machine learning models were tested using 408 observation data collected from a real production line, with the predictive performance rates achieved by the XGBoost ( R 2 = 0.799), Random Forest ( R 2 = 0.792), and Gradient Boosting ( R 2 = 0.777) algorithms. This study contributes to filling a gap in the literature regarding the modeling of sizing-related breakages using machine learning algorithms. Furthermore, it brings together data science and textile engineering by systematically presenting model comparisons and variable significance analyses. In practice, the study contributes to more effective decision-making aimed at reducing machine downtime and fabric losses by predicting the risk of breakage. The findings may also support sustainability-oriented manufacturing practices, particularly in terms of resource efficiency and reduced waste.