Machine learning-driven optimization of biphasic pretreatment conditions for enhanced lignocellulosic biomass fractionation

Biphasic pretreatment efficiently fractionates lignocellulosic biomass (LCB) and holds significant potential for industrial applications. While various studies have explored parameters to improve its efficiency, the lack of an optimal framework to balance these factors restricts scalability and compromises cost-effectiveness. This study introduces a machine learning (ML) model to optimize biphasic pretreatment conditions for LCB fractionation. By leveraging ML's capacity to uncover intricate relationships within extensive datasets, we conducted a comprehensive analysis incorporating key parameters. Feature importance analysis highlighted the critical influence of these parameters on cellulose degradation, hemicellulose removal, and delignification . The Gradient Boosted Regression (GBR) model outperformed others, achieving robust predictive metrics with R 2 values from 0.71 to 0.94 and demonstrating lower error levels (RMSE: 5.27–9.51; MAE : 3.73–7.49) compared to other models during validation. Solid loading and temperature were identified as the most influential factors, contributing 23.7 % and 21.3 % to cellulose degradation, respectively. For hemicellulose removal, solid loading accounted for 41.8 %, while temperature contributed 25.3 % to delignification. The GBR-based optimization achieved 10.7 % cellulose degradation, 98.9 % hemicellulose removal, and 91.2 % delignification, with relative errors of 5.6 %, −6.8 %, and −2.2 % upon experimental validation. This ML model can revolutionize optimizing processing conditions for LCB fractionation, significantly reducing experimental time and costs while enhancing bioenergy production efficiency. • Machine learning (ML) is applied to model and understand LCB fractionation. • Gradient boost regression model outperforms other models with R 2 0.71 to 0.94. • Solid loading and temperature are crucial factors in LCB fractionation. • Experimental validation confirms the ML model's high accuracy with low error.