Improving accuracy and interpretability of portable near-infrared spectroscopy via portable mass spectrometry fusion: A case study on the geographical traceability of Angelica sinensis

With growing consumer demand for food quality and safety, rapid and accurate origin traceability has become increasingly important. Portable near-infrared spectroscopy (NIR) and portable mass spectrometry (PMS) have shown respective advantages in food authenticity assessment. However, no study has yet explored the integration of both technologies for origin identification. This study proposes a multilevel data fusion strategy combining NIR and PMS to enhance the traceability accuracy of Angelica sinensis Radix (ASR), a regionally distinctive food ingredient. After applying the mahalanobis distance method to eliminate outliers, classification models based on random forest, radial basis function neural network, and serpent optimization-support vector machine (SO-SVM) were built using both NIR and PMS data. Preprocessing and variable selection techniques were employed to improve model robustness. The SG-MSC-2nd Der-SO-SVM model yielded the best performance on NIR data, while the SS-CMA-ES-SO-SVM model achieved optimal results for PMS data. Among fusion strategies, the Bayesian model averaging-based approach outperformed low-level and mid-level fusion methods, achieving 0.95 accuracy and a Kappa value of 0.91 on the independent validation set. This study demonstrates that the fusion of complementary spectroscopic and spectrometric data offers a powerful solution for improving origin traceability and provides a valuable reference for food quality monitoring.