RAPID DETERMINATION OF FAT CONTENT IN FISH BY NEAR-INFRARED SPECTROSCOPY COMBINED WITH PARTIAL LEAST SQUARES REGRESSION

Abstract

A fat content prediction model was built using near-infrared spectroscopy (NIR) combined with partial least squares regression (PLS). Twenty five samples of Nuc fishes were randomly collected for NIR measurements in 4 individual regions, and the fat content was also determined by chemical method. All data measured was used to develop and optimize the model. The model prediction was optimized based on the method of choosing significant wavelengths to obtain the highest R-square value and the smallest mean square error (MSE). The best prediction model was built based on NIR data in the lower abdomen of fish with the 0.96 of R-square and 0.001 of MSE for the cross-validation set.