2000c. Discrimination of Edible Oil Products and Quantitative Determination of Their Iodine Value by Fourier Transform Near-Infrared Spectroscopy

ABSTRACT: This work demonstrates the application of partial least squares (PLS) analysis as a discriminant as well as a quantitative tool in the analysis of edible fats and oils by Fourier transform near-infrared (FT-NIR) spectroscopy. Edible fats and oils provided by a processor were used to calibrate a FT-NIR spectrometer to discriminate between four oil formulations and to determine iodine value (IV). Samples were premelted and analyzed in glass vials maintained at 75°C to ensure that the samples remained liquid. PLS calibrations for the prediction of IV were derived for each oil type by using a subset of the samples provided as the PLS training set. For each oil formulation (type), discrimination criteria were established based on the IV range, spectral residual, and PLS factor scores output from the PLS calibration model. It was found that all four oil types could be clearly differentiated from each other, and all the validation samples, including a set of blind validation samples provided by the processor, were correctly classified. The PLS-predicted IV for the validation samples were in good agreement with the gas chromatography IV values provided by the processor. Comparable predictive accuracy was obtained from a calibration derived by combining samples of all four oil types in the training set as well as a global IV calibration supplied by the instrument manufacturer. The results of this study demonstrate that by combining the rapid and convenient analytical capabilities of FT-NIR spectroscopy with the discriminant and predictive power of PLS, one can both identify oil type as well as predict IV with a high degree of confidence. These combined capabilities provide processors with better control over their process. Paper no. J9230 in JAOCS 77, 29-36 (January 2000) KEY WORDS: Discriminant analysis, fats and oils, Fourier transform near-infrared spectroscopy, iodine value, partial least squares, quality control. The application of near-infrared (NIR) spectroscopy in edible oil analysis has predominantly involved its use for the rapid quantitative determination of the oil content in oilseeds, with relatively little work being carried out on the analysis of oils per se. Most work on NIR oil analysis development has focused on classifying and/or discriminating between oil types as well as detecting adulteration, particularly of olive oil. Bewig et al. (1) used a filter-based NIR instrument to differentiate between four types of oils (cottonseed, canola, soybean, and peanut) using discriminant analysis based on Mahalanobis distance principles. Sato (2) used principal component analysis (PCA) to classify vegetable oils using second-derivative NIR spectra, with PCA providing the benefit of using all the spectral data collected rather than only the data at selected wavelengths. Wesley et al. (3