Application Of Chemometric Algorithms In Quality Research Of Edible Vegetable Oils

Different kinds of vegetable oils possess different qualities because of their chemical compounds. In China, the price of extra virgin olive oil (EVOO) is more than several times to dozens of times higher than other ordinary vegetable oils. Obviously, common people cannot undertake so high cost of living, so exploring a kind of vegetable oil which has similar quality compared with EVOO has some practical significances. It can offer people more choice for vegetable oils. Besides, in order to satisfy the requirement for fast and real-time detection of vegetable oils, we establish a method for the measurement based on Raman spectrum.In this paper, after the vegetable oil samples were esterified, gas chromatography/mass spectrum (GC/MS) was used for the detection. Then the NIST library with equivalent chain length value (ECL) and internal standard method were adopted for the qualitative and quantitative analysis, respectively. After that, principal component analysis (PCA) and unsupervised random forests (RF) were applied to cluster the samples. RF showed a better ability of discrimination compared with PCA, because six kinds of vegetable oils belonged to six different zones. Meanwhile, plot of the RF proximity matrix demonstrated that tea oil was the closest to EVOO samples, therefore, tea oil was the most similar to EVOO from the perspective of fatty acid profiles. Besides, RF also provided the measure for variable importance, which allows us to assess the contribution of each variable for the clustering analysis. Results showed that palmitic, linoleic, linolenic, stearic and oleic were the most important variables for clustering the oils.In this paper, considering many shortcomings of chromatography method, such as expensive cost, time-consuming process and complex pretreatment, we established a rapid method for the determination of monounsaturated fatty acids (MUFA) and polyunsaturated fatty acids (PUFA) by Raman spectrum. At first, the Raman spectrum data was pretreated by smooth, background subtraction methods and normalization. Then, partial least square (PLS) was used for establishing the calibration model for the two targets. Prediction results showed that there was strong linear relationship between the relative content with the Raman spectrum. The correlation coefficient for the testing set (Q2) of MUFA and PUFA was0.9522and0.9944, respectively; root mean square error prediction (RMSEP) was0.2195and0.0673, respectively, which were all up to the standards of quantitative measurement. In conclusion, the rapid determination of the two important targets (MUFA and PUFA) using Raman spectrum is feasible.