Spectral Variable Selection for Partial Least Squares Calibration Applied to Authentication and Quantification of Extra Virgin Olive Oils Using Fourier Transform Raman Spectroscopy

摘要: The limits of quantitative multivariate assays for the analysis extra virgin olive oil samples from various Greek sites adulterated by sunflower have been evaluated based on their Fourier transform (FT) Raman spectra. Different strategies wavelength selection were tested calculating optimal partial least squares (PLS) models. Compared to full spectrum methods previously applied, optimum standard error prediction (SEP) concentrations in spiked could be significantly reduced. One efficient approach (PMMS, pair-wise minima and maxima selection) used a special variable strategy consideration significant respective PLS regression vectors, calculated broad spectral intervals low number factors. PMMS provided robust calibration models with small variables. On other hand, Tabu search recently published (search process guided restrictions leading list) achieved lower SEP values but at cost extensive computing time when searching global minimum less Robustness was using packages ten twenty randomly selected within cross-validation independent values. best one year's harvest total 66 Cretian obtained such optimized leave-20-out (values between 0.5 0.7% weight). For more complex population all over Greece (total 92 samples), results 0.7 0.9% weight sample package size 20. Notably, method has shown valid chemometric which single model can applied 1.4% across three different years.