113 Classification of Blended Edible Oils Based on Raman Spectroscopy with Chemometrics

Raman spectroscopy was utilized to distinguish the brands of edible oils. As there are more than four thousands of Raman spectral variables which can cause the discrimination model redundancy, it is necessary to do data mining. The 100.0% correct answer rate (CAR) result of principal component analysis (PCA) - least-square support vector machine (LS-SVM) model shows that it is feasible to distinguish different brands of blended edible oils using Raman spectra. However, the PCs used for the model establishment were not easy to be interpreted. Variable selections can let the model more and interpretable. The performance of successive projections algorithm (SPA)-LS-SVM model became compared to full-spectrum-LS-SVM model. Our results proved that uninformation variable elimination (UVE) can much eliminated uninformation variables, and it is necessary to operate UVE before SPA, which can both reduce the calculation time and increase the model's performance. As a conclusion, it is feasible to distinguish different brands of rapeseed oils using Raman spectroscopy, and SPA can do more effective variable selection according to UVE.