Rapid screening of fumonisins in maize using near-infrared spectroscopy (NIRS) and machine learning algorithms.

Fumonisins occurrence in maize represents a significant global challenge, impacting economic stability and food safety. This study evaluates the potential of near-infrared (NIR) spectroscopy combined with chemometric algorithms to detect fumonisins in maize. For fumonisin B1 (FB1) and B2 (FB2) levels were developed predictive NIR models using partial least squares (PLS) and artificial neural networks (ANN). PLS models demonstrated strong correlation coefficient (R) values of 0.90 (FB1), 0.98 (FB2), and 0.91 (FB1 + FB2) for calibration, with ratio of prediction to deviation (RPD) values ranging 2.8-3.6. Similarly, ANN models showed good predictive performance, particularly for FB1 + FB2, with  = 0.99, and the root means square error (RMSE) of 131 μg/kg for calibration; and  = 0.95, RMSE = 656 μg/kg for validation. These findings underscore the efficacy of NIR spectroscopy as a rapid, non-destructive tool for fumonisin screening in maize, with chemometric algorithms enhancing model accuracy, offering a valuable method for ensuring food safety.