Advanced non-destructive detection of peanut adulteration in ground roasted hazelnut using FT-NIR spectroscopy and machine learning

Hazelnut adulteration with ground peanut is a severe health and economic problem. In this study, Fourier transform near-infrared spectroscopy (FT-NIRS) was combined with two distinct machine learning (ML) algorithms (the Lasso and Elastic Net) to detect and quantify the peanut adulteration rate (%). Ground hazelnut samples were mixed with various levels of peanut content (0-50%). The spectral data were collected in the range of 4000-10000 cm^-1. Feature selection was carried out using the Lasso and Elastic Net algorithms to determine and eliminate unnecessary spectral variables and improve the accuracy of prediction. The Lasso model was found to be more accurate (R²_test = 0.93, RPD = 3.9, RMSEP = 3.7%) compared to the Elastic Net model (R²_test = 0.87, RPD = 2.9, RMSEP = 5.0%) for the same λ value (0.001). Moreover, the prediction accuracy indicator values improved as λ values decreased. Cross-validation confirmed the robustness of the Lasso model, indicating it is highly generalizable. The study findings indicated that FT-NIRS, supported by ML-based feature selection and modelling, provides an efficient, fast and non-destructive approach for the detection and quantification of hazelnut adulteration with ground peanut. This tested approach offers a rapid, waste-free, and eco-friendly solution to food adulteration detection, aligning with sustainable production principles by minimising sample preparation and resource consumption in the frame of greener analytical workflows.