Machine learning prediction of cellulose-rich materials from biomass pretreatment with ionic liquid solvents

Ionic liquid solvents (ILSs) have been effectively utilized in biomass pretreatment to produce cellulose-rich materials (CRMs). Predicting CRM properties and evaluating multi-dimensional relationships in this system are necessary but complicated. In this work, machine learning algorithms were applied to predict CRM properties in terms of cellulose enrichment factor (CEF) and solid recovery (SR), using 23-feature datasets from biomass characteristics, operating conditions, ILSs identities, and catalyst. Random forest algorithm was found to have the highest prediction accuracy with RMSE and R-2 of 0.22 and 0.94 for CEF, as well as 0.07 and 0.84 for SR, respectively. Highly influential features on making predictions were mainly from biomass characteristics and ILS treatment 's operating conditions, totally contributed 80% on CEF and 60% on SR. Oneand two-way partial dependence plots were used to explain/interpret the multi-dimensional relationships of the most important features. Our findings could be applied in designing new ILSs and optimizing the process conditions.

Automated summary

In this study, machine learning algorithms were used to predict CRM properties in terms of CEF and SR, with the random forest algorithm showing the highest prediction accuracy with R-2 values of 0.94 for CEF and 0.84 for SR. Biomass characteristics and ILS treatment operating conditions were found to be highly influential in making predictions. One-and two-way partial dependence plots were used to explain the multi-dimensional relationships of the most important features. Our findings could be applied in designing new ILSs and optimizing the process conditions.