XGBoost and LightGBM for Spectral Authentication of Pomegranate Juice Concentrate: A Comparison with Classical Chemometric Models

Ensuring the authenticity of fruit juice products is a growing concern, particularly for high-value pomegranate juice concentrate (PJC), which is vulnerable to economically motivated adulteration. This study evaluated Fourier-transform infrared (FTIR), UV–visible (UV–Vis), and fluorescence spectroscopy to detect adulteration of PJC with sugar syrups (invert and liquid) and fruit juice concentrates (apple and grape). A total of 166 authentic and adulterated samples were analyzed by each technique, and the resulting spectra were modeled using chemometric (PLS-DA, OPLS-DA, PLS, OPLS) and gradient boosting algorithms (XGBoost, LightGBM). Spectral analyses showed clear differences between authentic and adulterated samples. A sequential multi-level classification strategy yielded high diagnostic accuracy, with FTIR-based models giving the best performance and validation correct classification rates above 97%. For adulterant level prediction, FTIR-based PLS and XGBoost models achieved R2 values above 0.97 and RPD values above 7, indicating strong quantitative capability. UV–Vis and fluorescence models provided moderate performance, with fluorescence models showing the greatest variability. Overall, the findings highlight the superior performance of FTIR spectroscopy, particularly when coupled with gradient boosting, and support its use with robust modeling tools as a rapid, reliable approach for routine juice authenticity assessment in industrial and regulatory settings for screening and confirmation of adulteration.