Machine learning-assisted near-infrared spectroscopy for rapid discrimination of apricot kernels in ground almond

Abstract

Almonds are one of the most widely consumed seeds in the world, both for their taste and for their high nutritional value. A rapid and non-destructive method to detect adulteration of ground almond with apricot kernels is a necessity in the food industry because of almond's high commodity value and being one of the most consumed tree nuts. Almonds are a target for economically motivated adulteration, and apricot kernel is the most seen adulterant in ground almond. NIR spectroscopy is simple, non-destructive, and cheaper alternatives to traditional methods including chromatography for the detection of almond adulteration. A total of 120 almond samples were purchased in Türkiye. NIR spectra were collected using a portable and benchtop spectrometer and analyzed by Soft Independent Modeling of Class Analogy (SIMCA) and Conditional Entropy (CE) with machine learning algorithms to generate a classification model to authenticate ground almonds. Partial Least Square Regression (PLSR) and CE with machine learning algorithms were used to predict the levels of apricot kernel in ground almonds. Ground almonds were adulterated with apricot kernels at different level (0–50%) with 2% intervals. Both SIMCA and CE algorithms combined with spectral data obtained from the spectrometers provided very distinct clusters for pure and adulterated samples (100% accuracy). Both units also showed superior performance in predicting apricot kernels using PLSR with rval>0.96 with a standard error prediction (SEP) 3.98%. Besides, CE with machine learning algorithms reveal similar performance using benchtop NIR spectrometer (SEP>4.49). Based on the SIMCA, PLSR, and CE-based models, NIR spectroscopy can be used as an alternative methods and showed great potential for real-time surveillance to detect apricot kernel adulteration in ground almond.