Vis/NIR spectroscopy and chemometrics for enzymatic softening prediction in quince fruit

Browning and softening are critical physiological processes that affect fruit quality during postharvest storage, primarily influenced by enzymatic activity of polyphenol oxidase (PPO), peroxidase (POD), and pectin methyl-esterase (PME). Conventional enzyme activity assays are time-consuming, costly, and require chemical reagents. To address these limitations, this study employed Visible-Near Infrared (Vis/NIR) spectroscopy combined with advanced regression modeling to predict PPO, POD and PME activities in quince fruit non-destructively. Multiple linear regression (MLR), partial least squares regression (PLSR), and artificial neural network (ANN) were developed. All models provided reliable predictions (R² > 0.92, RPD > 2.8), demonstrating the method’s robustness. To enhance model interpretability and reduce computational complexity, a novel hybrid variable selection combining decision tree (DT) optimization with eleven metaheuristic algorithms was applied to identify the most informative wavelengths. Effective-wavelengths (EWs) selected by Learning Automata (LA) and Forest Optimization Algorithm (FOA) were chosen to improve regression algorithm performance. A comparable performance was observed between full-spectrum and EW–based models. Full-spectrum models benefit from broader chemical representativeness and potentially enhanced robustness, whereas EW-based models offer clear advantages in terms of reduced dimensionality, improved interpretability, and suitability for real-time or embedded sensing applications. The methodology offers broad applicability for on-line quality control and real-time monitoring of enzymatic browning and softening in other fruit commodities, supporting improved postharvest management and reduced food losses.