Detección de almidón en leche en polvo basado en espectroscopia Raman y mínimos cuadrados parciales

This research aimed to establish a mathematical model, using Raman spectral information and the partial least squares regression algorithm (PLS), to predict the percentage of adulteration of powdered milk by starch. The regression model obtained can be used to identify samples that show starch in powdered milk in concentrations ranging from 5% to 40% (w/w). The cross-validation method was used with the strategy of leaving a sample out. The interval that was optimal is the wave number range of 2170-2272 cm-1 . The linear regression model obtained has a multiple correlation coefficient of 99.99%, minimum sum of squares of the predicted residual error (PRESS) of 237.4 and the value of the F statistic, 19210.29 allows us to establish that if there is a relationship Linear significance between Raman intensities and the values of starch concentrations in the mixture. The value of the critical level p = 0.006 indicates that there is a significant linear relationship, and therefore, that the hyperplane defined by the regression equation offers a good fit.