Feature variable selection for identifying salmonids cultured in freshwater and seawater based on chemometrics analysis and data fusion strategies.

The selection of feature variables to distinguish between freshwater- and seawater-farmed salmonids is crucial for building reliable traceability methods. Here, salmonids were cultured for 94 days under three different salinity change regimes. Their stable isotopes, elements, and phospholipid fatty acids were characterized. The different variables exhibited different sensitivities to salinity changes. These data were integrated using various data fusion strategies to create five datasets with 40, 12, 12, 9, and 7 variables, respectively. The five datasets were coupled with support vector machine (SVM), random forest (RF), linear discriminant analysis (LDA), and orthogonal partial least squares-discriminant analysis (OPLS-DA) to differentiate 86 salmonids from different production methods. A satisfactory discrimination rate of 100 % was achieved with SVM and Dataset IV. The nine variables in this dataset (δH, δO, Sr, C18:0, ΣSFA, C20:3n3, C22:6n3, C18:2n6, and ΣPUFA) are promising indicators for discriminating between salmonids cultured in freshwater and seawater.