Abstract Background Effective modeling and parametric studies strengthen analyses of the influence of cotton fiber properties on the yarn properties, which are important for developing a higher-quality product. Machine learning (ML)-based predictive algorithms have been adopted for accurate modeling of fiber-yarn relationships. However, the complex black-box models lack interpretability and transparency, thereby limiting parametric studies. Therefore, an integrated ML-based black-box modeling and explainable artificial intelligence (XAI)-based analysis can facilitate accurate yet interpretable parametric analysis of yarn quality. Results In this study, three ML algorithm-based models, i.e., random forest, support vector regression, and K-nearest neighbors, were developed based on an experimental cotton fiber dataset. Five cotton fiber properties were considered input variables for predicting yarn tenacity and unevenness. Based on their performances, the most appropriate model for each yarn property was selected. Shapley additive explanations (SHAP), an XAI technique, is applied to provide interpretability to the models’ predictions and study the contributions of fiber properties. Short fiber content, with a mean absolute SHAP value of 0.359, was the most significant property, followed by fiber strength (0.242). Unevenness was influenced maximally by short fiber content, with a mean absolute SHAP value of 0.735. Other fiber properties exhibited comparatively weaker and nonlinear influences. Conclusions The study conducts an analysis of fiber-yarn relationships, aided by SHAP-based analysis. The parametric interpretations of the SHAP technique were validated using linear regression and parametric sweeping, strengthening the robustness of the deduced nature of relationships. The proposed framework provides a practical and interpretable decision-support tool for the analysis of the quality of yarn and the comprehension of its production through fiber processing.
Sarker et al. (Thu,) studied this question.