Desenvolvimento de metodologia portátil e não destrutiva para determinação de adulteração em leite de cabra

Abstract

This study proposes an innovative approach for determining fat content and detecting cow milk adulteration in goat milk samples using a miniaturized near-infrared (NIR) spectrometer combined with advanced multivariate calibration techniques. Statistical models based on the Successive Projections Algorithm (SPA) for variable and interval selection were applied through Multiple Linear Regression (SPA-MLR) and interval Partial Least Squares (iSPA-PLS) methods. Among various preprocessing techniques tested, the 11-point Savitzky-Golay smoothing (SGS) provided the best predictive performance for the quantitative models. The SGS/iSPA-PLS model achieved high correlation coefficients (r_pred) of 0.97 and 0.99, root mean square errors of prediction (RMSEP) of 0.12 g/100 g and 2.15 g/100 g, ratios of performance to deviation (RPD) of 4.32 and 8.96, and relative errors of prediction (REP) of 2.70% and 8.04%, respectively for fat content estimation and adulteration detection. This methodology effectively addresses challenges related to compositional variability and adulterants presence, providing a robust, portable, and efficient analytical tool suitable for scientific research and quality control of goat milk.