Soy Sauce Classification by Geographic Region Based on NIR Spectra and Chemometrics Pattern Recognition

ABSTRACT Statistical and artificial neural network (ANN) pattern recognition techniques were applied to NIR spectra of 38 soy sauce samples collected from the northern/central, western, and southern regions in Japan and related to differences in food flavorings. Linear discriminant analysis (LDA) and ANN using factor scores calculated from NIR spectra showed more accurate differentiations than those based on the original spectra. In LDA, the correctly assigned ratio was 81.6%. Correct classification ratios shown by Partial least squares (PLS2) were 84.2% and by ANN 76.3% in the cross‐validation test. The differentiations suggested that there are quality differences in soy sauce among the three regions in Japan.