Machine learning-guided inverse design of biomass-derived MgO nanoparticles with targeted sizes: A case study on As(III) adsorption.

Achieving precise size control in the synthesis of biomass-derived magnesium oxide (MgO) nanoparticles remains a significant challenge. Here, this study presents an inverse design framework integrating interpretable machine learning and multi-stage optimization for this purpose. Using experimental data on biomass categories, precursors, and process parameters, we evaluated 12 machine learning models. The five top-performing tree-based ensemble models were subsequently integrated via a stacking method, yielding an optimal model with an area under the receiver operating characteristic curve (AUC) of 0.9534 on the test set. The SHapley Additive exPlanations (SHAP) framework systematically revealed the differential and nonlinear fstatistical association patterns between various process parameters and the target size category, thus providing data-driven insights into the size evolution patterns observed during green synthesis. The constructed reverse design system successfully identified a synthesis formulation, while the experimentally produced MgO nanoparticles exhibited a particle size distribution of 61-88 nm, which was consistent with the targeted medium size range. The resulting material achieved a maximum As(III) adsorption capacity of 570.42 mg/g, which surpassed the capacity of commercial magnesium oxide. This preliminary finding underscores the application potential of materials synthesized via the proposed reverse design framework. The novelty of this work originates from the synergistic integration of interpretable machine learning and multi-stage optimization, providing a reliable and interpretable pathway for the inverse design of green synthesis processes from target dimensions. This methodology establishes a data-driven framework and a decision-support tool for the efficient and targeted synthesis of biomass-assisted nanomaterials.