Partial Least Squares Discriminant Analysis Model Based on Variable Selection Applied to Identify the Adulterated Olive Oil

The identification of the authenticity of edible vegetable oils is important from both consumer health and commercial aspect. Fourier transform infrared spectroscopy combined with multivariate statistical analysis methods was used to identify the authenticity of olive oils. Partial least squares discriminant analysis (PLS-DA) based on a reduced subset of variables was employed to build classification models. For the purpose of variable selection, a modified Monte Carlo uninformative variable elimination (MC-UVE) technique was proposed. Comparing with other variable selection techniques, PLS-DA model using the selected variables by the modified MC-UVE provided better results. The classification accuracy obtained by cross validation was 97.6 %, and the correct classification rate of the prediction set was 100 %. The results show that the model based on the modified MC-UVE is successful in the inspection of the authenticity of olive oils.