Rough set machine learning-based model for prediction of deep eutectic solvent pretreatment of lignocellulosic biomass
Phuong Anh T. Vuong, Nishanth G. Chemmangattuvalappil, Kiat Moon Lee, Jecksin Ooi
Abstract
Abstract Recently, deep eutectic solvents (DES) have shown promising results for its application in the pretreatment of lignocellulosic biomass (LCB) to produce fermentable sugars. The optimal pretreatment conditions have been determined through predictive models developed from machine learning algorithms. These models, however, are black box models where the results cannot reveal the scientific reasons behind the identified relationships. To address such limitations, this work employed rough set machine learning (RSML) to create an accurate and interpretable predictive model capable of analysing DES efficiency in pretreating LCB. RSML generates results represented as if–then rules by learning from a dataset composed of conditional and decision attributes. The selected conditional attributes are DES composition, LCB properties (including type and composition of biomass), and pretreatment conditions, while sugar yield is the decision attribute. The RSML model possesses 94.5% and 90.3% predictive ability when applied to validation set and testing set, respectively, recommending that to achieve sugar yield above 75%, high temperature (> 105 °C), low DES-to-biomass ratio (< 5.8), and short duration (< 2.25 h), and acid-based DES are required. Graphical Abstract