As the textile industry faces growing challenges related to sustainability, recycled fiber blending for making new yarns has emerged as a key area for reducing environmental impacts. This study aims to investigate the role of fiber length features in predicting the quality of blended yarns, particularly focusing on natural-based fiber blends such as recycled cotton (ReCo) and Lyocell. Machine learning models, including Random Forest, Gradient Boosting, and Support Vector Regression, alongside linear and polynomial regressions, are used to predict fiber properties based on empirical data. The results show fiber length features from the Staple Diagram and Fibrogram as the most significant factors. Hyperparameter tuning has enhanced model accuracy, especially for Random Forest and Gradient Boosting, showing significant reductions in error metrics. Cross-validation is performed to ensure the reliability of the models and prevent overfitting during the predictive analysis of fiber length features. Shapley Additive Explanations (SHAP) analysis reveals that specific fiber length ranges have the most influence on model predictions, highlighting their importance in optimizing blended yarn properties. These findings contribute to advancing sustainable textile production through data-driven approaches and textile fiber blend optimization.