Comparison of Fourier Transform near Infrared Spectroscopy Partial Least Square Regression Models for South African Extra Virgin Olive Oil Using Spectra Collected on Two Spectrophotometers at Different Resolutions and Path Lengths

Abstract

Near infrared (NIR) partial least square (PLS) regression models for the determination of several extra virgin olive oil quality parameters (peroxide value, free fatty acid content, specific extinction coefficients K232 and K270, linoleic, oleic and saturated acid content, α:β ratios of linoleic and oleic acid, pigment and total polyphenol content, thiobarbituric acid reactive substances value) were developed from spectra collected on two different Fourier transform near infrared (FT-NIR) spectrophotometers using different sample holders, path lengths and resolutions. Spectra were recorded on the PerkinElmer IdentiCheck spectrophotometer (9091–4000 cm−1) in transmittance mode at resolutions of 64, 32, 16 and 8 cm−1 at each of two path lengths (0.2 and 0.5 mm) and on the Büchi NIRLab N-200 instrument (10 224–4000 cm−1) in transflectance mode at a fixed resolution of 8 cm−1 and a path length of 0.6 mm. The PLS regression models, for each respective parameter, were statistically compared to evaluate the effect of different sample scanning conditions on model performance. The comparison of the regression models developed from PerkinElmer spectra revealed that spectra recorded at the lowest resolution of 64 cm−1 produced equally accurate models when compared to higher resolution spectra and that the two path lengths resulted in no significant differences. Comparisons between PLS models developed from Büchi and PerkinElmer spectra, respectively, showed significant differences for only a few parameters. In general, reliable prediction results were obtained for the peroxide value [standard error of prediction ( SEP) = 4.15 meq O2 kg−1, r2 = 0.87], K232 ( SEP = 0.94, r2 = 0.94), linoleic acid ( SEP = 0.83%, r2 = 0.90) and saturated acid content ( SEP = 0.91%, r2 = 0.88) PLS regression models.