Portable near infrared spectrometer to predict physicochemical properties in cape gooseberry (Physalis peruviana L.): An approach using hierarchical classification/regression modelling

Cape gooseberries are highly valued for their taste, nutraceutical benefits, and health properties, earning them recognition as a superfruit. However, these properties vary according to the ripening stage, making it important to monitor the composition of cape gooseberries throughout their maturation. In this study, we used a portable NIR spectrometer (900–1700 nm) combined with chemometrics to predict soluble solid content (SSC), vitamin C content, and firmness. 700 cape gooseberries in each of the four ripening stages (unripe, half-ripe, ripe, and overripe) were harvested from 2022 to 2023 at Bambamarca and Otuzco (Peru). Principal component analysis (PCA) revealed distinct clusters of cape gooseberries based on ripening stage, though no differences were observed between the seasons. Partial Least Squares Regression (PLSR) accurately predicted vitamin C content and SSC, with RMSEP values of 3.13 mg/g juice and 0.52 °Brix, respectively. The implementation of Competitive Adaptive Reweighted Sampling (CARS) and Bootstrapping Soft Shrinkage (BOSS) as variable selection methods improved RPD values by 4–7.6 %. PLSR was less effective at predicting firmness (N), particularly for unripe cape gooseberries. To address this, a hierarchical classification/prediction model was developed. In the first level, Partial Least Squares Discriminant Analysis (PLS-DA) successfully discriminated (error