Rapid Raman spectroscopy analysis assisted with machine learning: a case study on Radix Bupleuri.

Area of Science:

• Utilizes advanced spectroscopic techniques and computational methods for botanical analysis.
• Integrates machine learning for complex data interpretation in natural product chemistry.

Background:

• Radix Bupleuri is extensively used for its medicinal properties, but its efficacy is influenced by environmental factors affecting component concentrations.
• Variability in Radix Bupleuri necessitates robust methods for quality control and origin determination.

Purpose of the Study:

• To develop and evaluate machine learning models for classifying and predicting the production regions of Radix Bupleuri.
• To assess the performance of various machine learning algorithms, including LDA, SVM, XGBoost, and LightGBM, based on Raman spectral data.

Main Methods:

• Collected Radix Bupleuri samples from diverse regions and varieties.
• Acquired experimental and computational Raman spectra for sample analysis.
• Applied dimension reduction using Linear Discriminant Analysis (LDA) followed by classification and regression modeling with Support Vector Machine (SVM), XGBoost, and LightGBM.

Main Results:

• Support Vector Classifiers achieved a high accuracy of 98% and an F1 score above 0.96 in predicting production regions.
• Support Vector Regression demonstrated good fitting performance with an R² score above 0.90.
• Complex models exhibited a tendency towards overfitting, impacting generalization ability.

Conclusions:

• The LDA-SVM model proved to be a high-performing and stable method for Radix Bupleuri classification, aligning with High-Performance Liquid Chromatography (HPLC) results.
• This rapid classification and regression technique shows potential for application to other medicinal herbs, enhancing quality control and authentication.