Geographical origin traceability of chrysanthemum (Chrysanthemum morifolium) by fusion of NIR and ICP-MS enhanced by explainable XGBoost-SHAP model

Chrysanthemum is a widely cultivated medicinal and edible plant in China, known for its diverse potential health benefits. Significant differences exist in the bioactivity and commercial value of chrysanthemum from different geographical origins. Currently, there is a lack of objective and accurate methods for reliably tracing the geographical origin of chrysanthemum. Herein, we propose an interpretable machine learning strategy that integrates near-infrared spectroscopy and inductively coupled plasma mass spectrometry (ICP-MS) for geographical origin identification. A total of 100 independent chrysanthemum samples were collected from five representative production regions. The elemental profiles obtained from ICP-MS exhibit superior discriminative power compared to spectral information from NIR. Furthermore, the fusion of NIR and ICP-MS data provided a more comprehensive representation of sample characteristics, leading to significantly improved classification accuracy over models based on individual data types. Among the tested algorithms, the extreme gradient boosting (XGBoost) model on fused data achieved perfect classification accuracy (100%), outperforming both the linear PLS-DA and non-linear SVM models. SHapley Additive exPlanations (SHAP) analysis further highlighted the contribution of key elements, Sr, W, Ca, Si, Mo, Tb, Sn, Bi, Nd, and Cs, in differentiating geographical origins. Overall, our study presents a robust and interpretable framework for the geographical authentication of agricultural products. • NIR and ICP-MS were employed for tracing the geographical origin of chrysanthemum. • The ensemble XGBoost model significantly outperformed traditional PLS-DA and SVM. • The XGBoost classifier achieved accurate identification with accuracy of 100%. • SHAP analysis revealed key elements with substantial discriminative power. • A reliable tool was established for authenticity verification of agricultural products.