Machine learning-assisted Fourier transform infrared spectroscopy to predict adulteration in coriander powder

Abstract

Coriander is a widely used spice, valued for its flavor, aroma, and nutritional benefits in various cuisines and food products. However, adulteration, such as the addition of sawdust, poses significant risks to food safety and authenticity. This study aims to present a solution for predicting sawdust adulteration in coriander powder by providing a detailed methodology for utilizing machine learning-assisted FTIR spectroscopy. It employs various base models, including linear regression (LR), decision tree (DT), support vector regression (SVR), and artificial neural network, (ANN), for adulteration detection. It was observed that the original dataset and Savitzky–Golay smoothed dataset (dataset generated after preprocessing) yielded superior results by achieving R² values exceeding 0.92 and 0.96, respectively, for the validation set. It shows that more than 92 % of the variability observed in the adulteration detection is explained by the optimized ANN model due to complex non-linear relationship of adulteration level and spectral features. These findings highlight the potential of machine learning-assisted FTIR spectroscopy in accurately predicting sawdust adulteration in coriander powder. This offers promising prospects for enhancing food authentication practices by quantification of adulteration levels. The study also gives directions and methodology to quantify different types of adulterants in food products using machine learning-assisted FTIR spectroscopy, which can enhance food safety.