Abstract

The growing demand for sustainable chemical production has intensified interest in valorizing lignocellulosic biomass, particularly palm oil empty fruit bunches (EFBs), a major agro-industrial residue in Southeast Asia. Despite their high cellulose and lignin contents, EFBs remain underutilized due to their complex structure. This study investigated a green and efficient approach to EFB fractionation using deep eutectic solvents (DESs), with a focus on the choline chloride-lactic acid (CL) combination as the most effective. A novel aspect of this work is the application of a machine learning technique—Multi-Input Fuzzy Rules Emulated Networks (MiFRENs)—used here for the first time to optimize pretreatment conditions (temperature, time, liquid-to-solid ratio (SLR), and particle size) based on existing experimental data without requiring additional trials. The actual experiment conducted under MiFREN-predicted optimal conditions (141 °C, 1.87 h, LSR 35.71, and 40-mesh particle size) achieved 95.74 % lignin removal and 82.14 % cellulose purity. Structural characterization confirmed high-purity lignin composed mainly of guaiacyl (G) and syringyl (S) units. Subsequently, the pretreated EFB was hydrolyzed via microwave-assisted acid hydrolysis, producing levulinic acid (LA) up to 27.71 wt% based on cellulose. In addition, the CL-DES system had good reusability, with a 10 % decline in lignin removal efficiency after five cycles. This work established an integrated framework, combining green solvent chemistry and artificial intelligence-driven optimization and advancing the sustainable conversion of agricultural residues into high-value bio-based chemicals.