A Machine Learning Aided Yield Prediction Model for the Preparation of Cellulose Nanocrystals

Machine
learning is one of the most innovative tools that has entered
the materials science toolkit in recent years. This work employs a
machine learning strategy to develop a yield prediction model for
producing cellulose nanocrystals (CNCs). It analyses the critical
factors affecting the yield from CNCs by optimizing reaction conditions
and reducing experiments. First, a data set of CNCs is established,
including cellulose sources and reaction conditions. The Weighted
Average Ensemble (WAE) approach is applied to an ensemble of five
tree-based base models on the data set, and it was found that the
WAE surpasses all the base models. The impact of critical features
on yield prediction is analyzed with partial dependence plots and
individual conditional expectation plots. Batch experiments are mainly
used to produce CNCs, but these are time-consuming. In this context,
the WAE model is a promising tool for rapidly predicting the yield,
and this study provides an excellent gateway to improve the extraction
of CNCs with high yields.

机器学习近年来已成为材料科学领域中的一项极具创新性的工具。本研究采用机器学习策略，旨在构建一种用于生产纤维素纳米晶体（CNCs）的产量预测模型。通过优化反应条件并减少实验次数，该研究分析了影响CNCs产量的关键因素。首先，建立了一个包括纤维素来源和反应条件的CNCs数据集。在数据集上，应用了加权平均集成（WAE）方法对五个基于树的基模型进行集成，结果表明WAE方法超越了所有基模型。通过局部依赖图和个体条件期望图分析了关键特征对产量预测的影响。批量实验主要用于生产CNCs，但这一过程耗时较长。在此背景下，WAE模型成为快速预测产量的有前景工具，本研究为提高高产量CNCs的提取提供了绝佳的切入点。