Traceability of wild Paris polyphylla Smith var. yunnanensis based on data fusion strategy of FT-MIR and UV-Vis combined with SVM and random forest

Abstract

Paris polyphylla Smith var. yunnanensis (Franch.) Hand.-Mazz (PPY) was a frequently used herbal medicine in pharmaceutical field and different provenances might affect the clinical efficacy. Tracing the geographical origin was an important portion for PPY authentication and quality assessment. Present study was compared low-, mid- and high-level data fusion methodology for geographical traceability of PPY samples (161 batches) combined with multivariate classification methods such as support vector machine gird search (SVM-GS) and random forest (RF) on the basis of Fourier transform mid-infrared (FT-MIR) and ultraviolet-visible (UV-Vis) spectra. Compared with the low- and mid-level data fusion strategy results basing on SVM-GS algorithm, result of high-level data fusion method (calculated by RF) was more satisfying. Result of RF basing on high-level data fusion strategy showed that merely two samples were misclassified and one sample was multiple assigned after voting with fuzzy set theory. Values of specificity, sensitivity, and accuracy rates were exceeded 0.91, 0.99 and 90.91%, for each class respectively, satisfying results of these were shown in training and test sets for high-level data fusion method. This feasible result indicated that the RF algorithm could establish a reliable and good performance model in geographical traceability on the basis of high-level data fusion strategy. Combination of high-level data fusion and RF algorithm could consider as a good choice for establishing a discrimination multivariate model for origins identification of PPY samples.

Keywords: Data fusion; Fourier transform mid infrared (FT-MIR); Paris polyphylla Smith var. yunnanensis; Random forest; Support vector machine gird search (SVM-GS); Ultraviolet-visible (UV–Vis).