Transfer Learning-Enabled Data-Driven Extraction of Cellulose Nanocrystals: Linking Cellulose Sources and Processing Conditions to Crystallinity

Transfer learning has emerged as a powerful tool in materials science, enabling accelerated materials discovery and design. Driven by increasing environmental concerns and the depletion of fossil-based resources, cellulose nanocrystals (CNCs) have gained significant attention as renewable and biodegradable materials. However, their extraction remains resource and energy intensive, necessitating efficient optimization strategies. Herein, a transfer learning approach is employed to address data-scarcity challenges and optimize the processing conditions of CNCs. A model trained for yield prediction is utilized as a base model for crystallinity prediction, leveraging a shared set of input descriptors, including cellulose sources and reaction conditions. This strong task relatedness enables effective knowledge transfer and improves predictive capability under data-scarce conditions. The yield prediction model achieves a root-mean-square error (RMSE) of 2.84 and a coefficient of determination ( R 2 ) of 0.84, while the transfer learned crystallinity model attains an RMSE of 2.94 and an R 2 of 0.83. By enabling rapid and resource-efficient screening of CNC extraction conditions, this framework reduces experimental trials, energy consumption, and material usage. Overall, this study demonstrates the potential of transfer learning-driven approaches to support sustainable materials processing and accelerate the development of environmentally benign materials.