Non-destructive origin traceability of functional food raw materials using portable near-infrared spectroscopy and variable selection-random forest algorithm

Based on portable near-infrared (NIR) spectroscopy and chemometric methods, a non-destructive strategy for origin traceability of functional food raw materials (FFRMs) was proposed. With Polygonatum odoratum , Taraxacum mongolicum , and Rheum palmatum as study subjects, spectra from different origins were collected using a grating portable NIR spectrometer at three integration times. Principal component analysis (PCA) and random forest (RF) methods were employed to construct the origin traceability models, while two variable selection-RF algorithms were developed to further enhance the accuracy. The performances were evaluated using the validation set and the prediction set collected one month later. The results indicated that optimization of integration time may improve performance in some cases. Furthermore, randomization test (RT) method demonstrated higher robustness and accuracy, while successive projections algorithm (SPA) method could simplify the RF model. 100% traceability accuracies were obtained with Rheum palmatum -RT/SPA-RF and Taraxacum mongolicum -RT-RF models, while over 95% accuracies were obtained with Polygonatum odoratum -RT-RF models. The variable selection-RF strategy provided a new approach for non-destructive origin traceability of FFRMs with high precision and robustness.