Data-driven model for the prediction of syngas composition and end-use application from gasification of residual biomass

Residual biomass gasification is a promising pathway for sustainable energy and chemicals' production. However, biomass feedstock variability and complex operational interactions limit the large-scale application of biomass gasification. This study presents a data-driven model to predict syngas composition and end-use application based on biomass’ properties and operational conditions, covering a wide range of residual biomass feedstocks, gasification conditions and reactor configurations. A CatBoost-based Decision Tree Regressor was trained on 133 experimental samples compiled from peer-reviewed studies. The dataset was augmented using the Synthetic Minority Over-sampling Technique (SMOTE) to mitigate class imbalance and enhance generalization across diverse gasification outcomes. Model performance was rigorously evaluated through nested cross-validation, yielding mean average errors of 4.2 ± 0.3 vol% for H 2 , 2.9 ± 0.4 vol% for CO, and 1.4 ± 0.3 vol% for CH 4 . Beyond composition predictions, the model accurately classified the most suitable syngas applications, achieving over 90 % identification for heat/power generation, methanol and biofuel synthesis, and SNG production. Moreover, the model was used for screening the differences in end use application of six different types of biomass residues. Steam gasification of sawdust and miscanthus grass emerged as particularly promising pathways for hydrogen-rich syngas generation, due to favorable combination of feedstock properties and reaction conditions. • Model predicts syngas composition and end-use based on biomass and process parameters. • Tree-based models provide interpretable pathways to understand predicted results. • Gasifying agent, volatiles, fixed carbon, moisture, and hydrogen are identified as key parameters. • Optimal H 2 yields require FC/VM < 0.2, H/C > 0.12, O/C < 0.85, moisture<20 wt% and alkali-rich ash.