The textile industry is confronted with significant sustainability challenges, particularly in recycling post-consumer cellulosic materials while preserving quality. Achieving an effective blend of a higher proportion of recycled fibres with pristine cellulosic fibres is essential for improving the quality and performance of textile products while reducing environmental impact. This research seeks to utilize machine learning techniques to optimize the blending process, ensuring that the resulting yarns adhere to industry standards.

This study adopts a data-centric approach, employing machine learning algorithms to analyse the properties of both recycled and pristine cellulosic fibres. Key attributes such as fibre length, strength, and fineness are derived from comprehensive datasets. Various regression models, including Random Forest and Gradient Boosting, are trained to predict the optimal blending ratios that produce desirable yarn characteristics. Hyperparameter tuning is performed to enhance model accuracy, and cross-validation techniques are used to ensure robustness.

The expected outcome is a predictive model that accurately forecasts the properties of blended fibres, aiding in the development of high-quality yarns. This research is innovative in its application of machine learning to the textile recycling process, offering a systematic framework for optimizing fibre blends based on empirical data. Future work will involve validating the model predictions against practical lab results to further refine the blending process. Additionally, the integration of SHAP (SHapley Additive exPlanations) analysis will be explored to improve interpretability and provide insights into feature contributions, guiding future research directions.