Data-driven prediction and optimization of nanocellulose yield in deep eutectic solvent

Nanocellulose possesses exceptional mechanical, functional, and environmental properties, while its sustainable production from biomass is limited by the complexity of solvent systems and processing conditions. Deep eutectic solvents (DES) offer a promising green alternative to traditional pretreatments. However, there is still a lack of systematic understanding of the factors that determine yield and the optimal operating conditions. This study introduces an integrated machine learning and optimization framework to predict and optimize the nanocellulose yield from DES-treated biomass. A curated dataset of 367 experimental samples from 34 studies was used to train an XGBoost model, which outperformed linear regression and neural networks in terms of root mean squared error (RMSE), mean absolute error (MAE), and R-squared. Model interpretation using SHAP identified mechanical treatment, temperature, feedstock purity, and primary HBD chemistry as the key factors influencing yield. The study also determined optimal ranges for feedstock ratios and reaction temperature. By integrating the trained model with the NSGA-II algorithm, a set of Pareto-optimal conditions was identified that balance high yield with reduced processing time. These optimal feature values align with established physicochemical knowledge and offer decision support for accelerating green nanocellulose production. This work demonstrates the potential of AI-driven methods to transform static empirical datasets into predictive and decision-support tools for sustainable materials engineering.