Rapid and non-destructive quantification of glucose and sucrose in Citrus aurantium L. infusion using laser-induced fluorescence and machine learning.

Bitter orange (Citrus aurantium L.) blossoms are rich in bioactive compounds and are widely valued for their medicinal properties. In this study, laser-induced fluorescence (LIF) spectroscopy was coupled with advanced machine learning to indirectly estimate glucose and sucrose concentrations in dried orange blossom (DOB) infusions. Nine concentrations (0-20 mg/mL) were prepared for each sugar in a 50 mg/mL DOB matrix and analyzed under 405 nm excitation across the 420-800 nm range. Fourier transform infrared (FT-IR) spectroscopy confirmed non-covalent sugar-phenolic interactions, with sucrose forming stronger hydrogen-bond networks than glucose, consistent with its more pronounced static fluorescence quenching. The minimum and maximum limits of detection for both sugars were 0.0321 mg/mL and 0.3320 mg/mL, respectively. Principal component analysis (PCA) revealed a clear separation between the sugar types but showed less discernible concentration gradients. This finding was reinforced by UMAP visualization and feature importance analysis, which highlighted the primary fluorescence emission region as the most influential. Classification analysis successfully distinguished sugar types and concentration trends, while support vector regression (SVR) models, validated using nested cross-validation, yielded R2 values of 0.9150 for sucrose, 0.8604 for combined sugars, and 0.7167 for glucose. The corresponding RPD values classified the sucrose model as excellent, the combined model as good, and the glucose model as unsuitable for quantitative use. These findings demonstrate that LIF combined with chemometric modeling can sensitively resolve subtle molecular differences in compositionally complex botanical systems, enabling fast and in situ sugar quantification for applications in food quality control, authenticity verification, and pharmaceutical monitoring.