Spectral Profile of Milk and Dairy Products: Methods of Mathematical Processing

Abstract

Dairy quality control requires rapid and relible analytical methods. Raman spectroscopy is time-saving, sensitive to molecular modifications, and reproducible. However, the resulting data are difficult to interpret due to complex multidimensional spectra, noise, and overlapping bands. These limitations are typical of multicomponent organic systems, such as milk and dairy products. A wide range of mathematical approaches and processing tools is required to analyze the primary spectra, classify them, model quantitative physicochemical indicators, and select optimal processing parameters. Nevertheless, no unified recommendations exist for the selection of processing and interpretation methods. This paper summarizes modern approaches to the mathematical processing of spectral data, including preprocessing methods, chemometric algorithms, and machine learning techniques. To improve the information content and comparability of the spectra, an effective analysis strategy involves comprehensive preprocessing, including artifact removal, baseline correction, smoothing, normalization, and scattering correction. Supervised learning methods are especially effective for modeling and prediction, notably partial least squares (PLS), as well as machine learning algorithms that account for complex nonlinear relationships. Integrated preprocessing and multivariate analysis improve the accuracy, reproducibility, and reliability of rapid dairy analysis.