LOW-COST "GREEN" ANALYTICAL STRATEGY BASED ON NEAR-INFRARED SPECTROSCOPY (NIRS) FOR DETERMINATION OF THE CHEMICAL COMPOSITION OF GOAT MILK

Analytical methods resulting from the use of NIR spectroscopy have characteristics such as speed, maintenance of sample integrity and absence of waste generation, which make them advantageous and viable for on-line industrial application.The use of portable NIR equipment as a low-cost alternative analytical tool for in situ characterization of food matrices represents an advance in the simplification and accessibility of technologies for small and medium-sized rural producers.   Thus, this work evaluated the suitability of the use of portable NIR equipment to determine the physicochemical composition of goat milk. The goat milk samples of the Anglo-Nubian (AN), Toggenburg (T), Mestizo (M) and Saanen (S) breeds were obtained between the months of May and October 2024 in the municipality of Soledade – PB. The reference analyses were carried out according to the Adolfo Lutz Institute, by the official methods: Kjedahl for proteins, Gerber for lipids, vacuum oven drying for moisture/total solids and carbohydrates for difference. The NIR spectra were obtained in duplicate in a portable NIR spectrometer (900 to 1700 nm), model NIR-S-T2 (InnoSpectra brand) operating in absorbance mode. Data treatment was performed using the Savitzy-Golay pre-processing chemometric technique with an 11-point window and 2nd order polynomial and by Partial Least Squares Regression (PLS). For the calibration and prediction sets, 30 and 12 samples were used, respectively. The PLS models for the calibration set of the different parameters showed the following results for Root Mean Square Error (RMSE) (g.100g⁻¹), Relative Error of Prediction (REP) (%), and correlation coefficient (R²), respectively: Total Solids (0.59; 4.34; 0.96); Lipids (0.44; 8.6; 0.98); Proteins (0.37; 11.44; 0.64); Carbohydrates (0.1; 2.18; 0.96). For the prediction set, the following results were obtained for RMSE (g.100g⁻¹), REP (%), and R², respectively: Total Solids (0.66; 4.85; 0.94); Lipids (0.74; 14.43; 0.92); Proteins (0.32; 9.62; 0.79); Carbohydrates (0.15; 3.35; 0.87). The accuracy of the model was assessed through the Elliptical Joint Confidence Region (EJCR) test in the prediction set. The results showed no statistically significant differences between the predicted values and the reference values, demonstrating that the predictive capacity of the PLS model is satisfactory. Therefore, it is concluded that the use of portable NIR in the physicochemical characterization of goat milk is promising, although requiring a larger number of samples as well as further chemometric adjustments in data analysis.