Utilizing a SHAP-Assisted Machine Learning Algorithm for Optimization of Plasmonic Properties of Gold Nanostructure

Gold nanostructures (GNSs) have tunable optical absorption and catalytic properties and thus are increasingly used in molecular sensing and nanomanufacturing. However, the rapid design of GNSs with specific optical properties is limited by various factors, making it difficult to achieve efficient design by conventional trial-and-error methods. Here, we report a SHapley Additive exPlanations (SHAP)-assisted extreme gradient boosting (XG-Boost) model for conducting the preparation of GNSs with preferential plasmonic features using silver nitrate (AgNO₃) and ascorbic acid in different surfactants at room temperature. The XG-Boost model achieves exceptional predictive accuracy (R² > 0.98, RMSE < 0.03). Importantly, SHAP elucidates the underlying mechanism linking synthetic conditions to plasmonic properties of GNSs, particularly revealing the contribution of surfactants to predictions, which was previously underestimated. Guided by the XG-Boost-SHAP strategy, GNSs with tunable localized surface plasmon resonance peaks and high stability are designed and precisely synthesized in various shapes. The GNSs were successfully applied to thiourea detection, Hg²⁺ recognition, glucose analysis, and fabrication of luminescent chips. The results demonstrate the remarkable potential of the machine-learning framework in tailoring GNSs with personalized features to realize various objectives.