Determination of the peroxide value in extra virgin olive oils through the application of the stepwise orthogonalisation of predictors to mid-infrared spectra

The potential of stepwise orthogonalisation of predictors (SELECT) coupled with multivariate calibration methods and pre-processing tools was examined to develop reliable and parsimonious regression models based on Fourier-transform mid-infrared (FT-MIR) spectroscopy for the prediction of peroxide value in extra virgin olive oils. The comprehensive calibration methodology proposed also involved the study of the structure of noise present in data, the preliminary detection of anomalous objects, and the appropriate construction of calibration sets. The improvement achieved through the application of feature selection was particularly remarkable when SELECT-Ordinary Least Squares (OLS) was applied on first-derivative spectra. In this case, a high-quality OLS model was obtained, providing a predictive ability similar to that achieved by full spectrum approaches (after noise removal and signal pre-processing) and with no significant signs of over-fitting (4.9, 5.3 and 5.2% root-mean-square errors of the residuals obtained in calibration, cross-validation and external prediction, respectively), but considerably improved properties in terms of model parsimony and interpretability. In fact, it should be noted that only 12 relevant predictors from a total of 408 original wavenumbers were selected to model peroxide value, which meant a very notable compression rate. The quality of the results obtained encourages the feasibility of implementing a FT-MIR based calibration strategy similar to that proposed in routine analysis for the assessment of other critical quality parameters for olive oil.