Integrated machine learning models for prediction of multiphase products from hydrothermal treatment of biomass

The lack of comprehensive tools for estimating product yields—bio-oil, bio-char, and bio-gas—from hydrothermal treatment (HTT) remains a major challenge for process optimization. To address this, this study introduces an Extreme Gradient Boosting (XGB) model to predict product yields from biomass HTT. A comprehensive dataset covering diverse feedstock compositions and operational parameters, including temperature, pressure, reaction time, and elemental ratios (C, H, N, O, S), was utilized. Input features were normalized, and hyperparameters were optimized via 10-fold cross-validation and full-factor grid search to ensure robustness. The XGB model achieved high predictive accuracy, with R2 values of >0.99 and 0.76–0.86 for training and test data, respectively. Model interpretability was enhanced using SHapley Additive exPlanations (SHAP), providing insights into feature importance. Key influential factors identified included temperature, carbon content, and residence time. Additionally, SHAP dependence and force plots offered an in-depth understanding of variable interactions. Overall, the developed model demonstrates potential for guiding experimental design and reducing trial-and-error in biomass conversion research. Multi-HTT, a user-friendly web-based interface, was developed to enable practical application of the model. Although limited by the lack of catalyst data and specific feedstock diversity, addressing these limitations in future work would improve model generalizability.