Predicting aflatoxin contamination in white and yellow maize using Vis/NIR spectroscopy combined with PCA-LDA and PLSR models through aquaphotomics approaches

• High classification accuracies for aflatoxin detection regardless of maize type. • PCA-LDA resulted 100% classification accuracies when models developed by type of maize. • PLS regression yielded high-performance models for aflatoxin quantification. • Water matrix coordinates revealed could be related to aflatoxin presence in maize. • NIRS can be an alternative approach for aflatoxin detection and quantification in maize. Conventional methods for the detection of aflatoxin require sample preparation and long analytical time. Researchers are therefore exploring cheaper, faster but reliable alternatives such as near-infrared spectroscopy (NIRS), which does not destroy the integrity of the food but rather, supports possible on-spot data driven decision making. This study aimed to develop models, optimized with pre-processing techniques and wavelength ranges to classify and predict 0, 3, 5, 10, 20, 30 and 50 ng/g aflatoxin in three major datasets (naturally contaminated white, spiked white maize and spiked yellow maize). Absorption peaks and bands (500, 950,1000, 1300, 1500, 1900, 2100, and 2300nm) were observed in the spectra, that could be related to aflatoxin contamination. Using all three datasets, the highest classification accuracies of 92.52% and 92.54% were obtained when models were developed at the wavelength range of 450-1050nm and1150-2400nm with Savitsky Golay smoothing (first derivative with filter 17). Sensitivity, precision, specificity and F1 score close to 1. Classification accuracies were 100% at all the distinct wavelength ranges when models were developed separately for each dataset. Partial least squares regression yielded an R 2 CV of 0.99, RMSECV of 1.70 ng/g, RPD of 9.90, LOD of 0.60 ng/g, LOQ of 1.81 ng/g at the wavelength range of 450-1050nm, indicative of model robustness and high-performance. Aquagrams revealed water matrix coordinates that could be related to aflatoxin presence in maize. The findings suggest that NIRS can be explored as a potential alternative approach for aflatoxin detection and quantification in maize.