Laboratory-Scale Feasibility of Fluorescence Spectroscopy for Detecting Cow Milk Adulteration in Plant-Based Milk Alternatives: Almond and Oat as Model Matrices

Cow milk adulteration in plant-based milk alternatives (PBMAs) raises authenticity and safety concerns, particularly for consumers with dairy allergies or lactose intolerance. This study evaluated the laboratory-scale feasibility of excitation–emission matrix (EEM) fluorescence spectroscopy combined with chemometric and machine-learning approaches for detecting and quantifying cow milk adulteration in almond and oat milk alternatives used as model matrices. Three independent preparation batches were produced for each PBMA matrix using one almond source, one oat source, and one commercial cow milk product, with cow milk concentrations ranging from 0 to 100% (v/v) and 2.5% as the lowest non-zero adulteration level. Parallel factor analysis identified 270 and 350 nm as informative excitation wavelengths, and the corresponding emission profiles were analyzed using principal component analysis, data-driven soft independent modeling of class analogy (DD-SIMCA), partial least squares regression (PLSR), random forest regression (RFR), and a one-dimensional convolutional neural network (1D-CNN). DD-SIMCA effectively rejected most adulterated samples, although target-class sensitivity was based on resubstitution because only three authentic spectra were available per condition. Under batch-grouped cross-validation, the best performance within the 0–50% adulteration range was obtained by PLSR for oat milk at 350 nm (RGCV2=0.877, RMSEGCV=6.12%, and RPDGCV=2.91), while RFR showed more consistent performance across matrix–wavelength combinations. The 1D-CNN exhibited greater variability and did not consistently outperform conventional models. These findings support the laboratory-scale feasibility of fluorescence-based screening for the investigated almond and oat matrices, while further validation using additional product sources, independently prepared batches, commercial samples, and lower adulteration levels is required before broader application to PBMA products can be established.