Rapid origin traceability and multi-element quantification of Hypericum perforatum L. using LIBS combined with machine learning methods

Abstract

Elemental analysis is crucial for determining the geographical origin of medicinal plants. This study employs laser-induced breakdown spectroscopy (LIBS) and machine learning to trace the origins and predict elemental content in Hypericum perforatum L. (HPL). LIBS data were collected from 269 HPL samples across various regions, while inductively coupled plasma mass spectrometry quantified 15 elements. Eleven preprocessing methods were evaluated for their impact on models. We developed four origin traceability models and multi-element quantification models using a multi-output regression framework. Optimal spectral intervals were identified through a moving window algorithm, and bootstrap aggregating enhanced model accuracy. The 4–8 interval effectively distinguished samples from different origins, with the XGBoost model performing particularly well in identifying July-harvested HPL from Xinjiang. All models achieved R² values exceeding 0.8574, and paired t-tests showed no significant differences between actual and predicted values, confirming their effectiveness in origin identification and elemental content prediction.