Accurate identification of goat milk on small-scale geographical origin: A nuclear magnetic resonance spectrometry and machine learning study.

Fuping goat milk powder is a kind of geographical indication protected food in China. To realize a rapid and accurate identification of Fuping goat milk, in this paper, an identification method was proposed based on nuclear magnetic resonance and machine learning to accurately distinguish Fuping goat milk on small scale geographical authenticity. Deuterated chloroform was selected as extraction solvent, partial least squares discriminant analysis (PLSDA), random forest (RF) and support vector machine (SVM) models were constructed using seven different geographical origin goat milk samples in Shaanxi Province. The identification based on PLS-DA model was unsatisfactory, the RF model performed better, and the SVM model showed best performance with accuracy rate reached 100% based on selected 17 features, and the 5-fold cross-validation accuracy was 94.7% ± 7.2%. It is disclosed that different geographic samples have different chemical compositions, particularly unsaturated fatty acids, such as conjugated linoleic acid.