Prediction and analysis of preparation of cellulose nanocrystals with machine learning

Extraction of cellulose nanocrystals (CNCs) from diverse cellulose sources is a promising and sustainable approach to produce nanocomposites. However, the traditional batch experiments are time/labor-consuming. Hence, three machine learning (ML) algorithms, i.e., decision regression tree, random forest, and artificial neural networks were applied to develop ML models. The dataset collected from published literature was used to train the ML models applicable to a wide range of cellulose source and reaction conditions. Among the three ML models, the random forest algorithm was the best one ( R 2 = 0.89, RMSE = 5.52) for the yield prediction, and the decision regression tree provided the highest accuracy ( R 2 = 0.86, RSME = 6.03) for the crystallinity prediction. The concentration of reagent and cellulose source was identified as the most important feature in yield and crystallinity prediction, respectively. The partial dependence analysis showed the impact of each input feature and their combined effects on the yield and crystallinity. This study may provide new perspectives and opportunities to understand and improve the preparation of CNCs.