Integration of high-throughput fluorescence and UV-vis spectroscopy for goat milk adulteration detection: comparison of one-class, binary classification and regression models

• UV-vis and fluorescence spectroscopy were integrated to detect goat milk adulteration. • One-class (SIMCA) and binary (PLS-DA, SVM) models were systematically compared. • Optimal SVM with fused spectra achieved 99.31% classification accuracy. • SVR with data fusion accurately quantified goat milk adulteration down to 5%. Adulteration of goat milk with cow milk has become an increasingly prevalent issue, highlighting the urgent need for rapid, non-destructive, and reliable authentication methods. In this study, ultraviolet-visible (UV-vis) and fluorescence spectroscopic fingerprints were combined with machine learning approaches and low-level data fusion to discriminate and quantify goat milk adulteration. Principal component analysis (PCA) was first employed to visualize sample distribution and explore data structure. Subsequently, one-class classification (SIMCA) and two-class classification models (PLS-DA and SVM) were established to identify goat milk adulteration. Among the tested models, the SVM model combined with low-level data fusion showed the highest overall performance, with an accuracy of 99.31%, sensitivity of 98.82%, and specificity of 100.00%, as well as PPV and NPV values of 100.00% and 98.36%, respectively. For quantitative analysis, partial least squares regression (PLSR) and support vector regression (SVR) were constructed, with the SVR model based on low-level data fusion providing the highest prediction performance ( R P 2 =0.9597, RMSEP = 3.6239, RPD = 4.49). Notably, this strategy enabled detection and quantification of adulteration with as low as 5% cow milk. Overall, the integration of UV-vis and fluorescence spectroscopy with low-level data fusion offers a rapid, green, and cost-effective approach for goat milk authentication.