Dynamic monitoring of salicylic acid and vitamin B2 during the apple boiling process based on fluorescence spectroscopy and machine learning.

Thermal processing affects nutrient retention in fruits, yet quantitative in-process monitoring remains challenging. Excitation-emission matrix (EEM) fluorescence combined with principal component analysis (PCA) was used to track changes in nutritional components during apple boiling. Overlapped fluorescence bands were resolved using Gaussian multi-peak empirical function fitting to derive peak descriptors. re regions were screened using minimum redundancy-maximum relevance (mRMR) and χ 2 tests; the synthetic minority over-sampling technique (SMOTE) was used for class balancing. Support vector machine (SVM) achieved 90.84-97.76% accuracy for boiling-stage classification ( κ = 0.9019-0.9761). Partial least squares (PLS) provided the best concentration prediction for both vitamin B 2 ( R 2 = 0 . 9734 , root mean square error (RMSE) = 0.0239) and salicylic acid ( R 2 = 0 . 9614 , RMSE = 0.0187). Kinetics showed distinct maxima: salicylic acid (11.588-28.796 ng/mL) peaked at 30-35 min (75-88 °C), while vitamin B 2 (2.164-6.001 ng/mL) peaked at 54-58 min at 100 °C. This peak-shape-aware EEM machine-learning framework supports non-destructive, quantitative process monitoring for nutrient retention during thermal processing.