Pistachio powder authenticity assessment using supervised machine learning algorithms

Adulteration of premium pistachio products with cheaper legume powders poses a significant food safety and economic concern. This study developed a non-destructive detection framework using visible-shortwave near infra-red (Vis-SWNIR) hyperspectral spectroscopy (400-1000 nm) combined with three machine learning algorithms-Discriminant Analysis (DA), Support Vector Machine (SVM), and K-Nearest Neighbours (KNN)-to discriminate pure pistachio powder from powder adulterated with 25% and 50% fava bean powder. Spectral absorption analysis revealed a monotonic decrease in absorption across all wavelengths (425-990 nm) with increasing fava bean concentration, with no wavelength showing higher absorption in adulterated samples. To reduce dimensionality and collinearity, Principal Component Analysis (PCA) was performed on the pre-processed spectra, and the first five principal components were retained as inputs for classification, explaining at least 99% of the cumulative spectral variance. Among the classifiers, DA achieved the highest accuracy (97.3%), followed by SVM (94.6%) and KNN (91.9%). All three classifiers perfectly or near-perfectly distinguished pure pistachio from adulterated samples, with errors occurring primarily between the two adulteration levels. The results demonstrate that Vis-SWNIR hyperspectral spectroscopy combined with DA or SVM provides a rapid, accurate, and non-destructive solution for detecting fava bean adulteration in pistachio powder, with potential for development into low-cost portable screening devices.