Aflatoxin B1, a major global food safety concern, is produced by toxigenic fungi during crop growing, drying, and storage, and shows increasing annual prevalence. This study aimed to detect aflatoxin B1 in chili samples using ATR–FTIR coupled with machine learning algorithms. We found that 83.6% of the chili powder samples were contaminated with Aspergillus and Penicillium species, with aflatoxin B1 levels ranging from 7.63 to 44.32 µg/kg. ATR–FTIR spectroscopy in the fingerprint region (1800−400 cm-1) showed peak intensity variation in the bands at 1587, 1393, and 1038 cm-1, which are mostly related to aflatoxin B1 structure. The PCA plots from samples with different trace amounts of aflatoxin B1 could not be separated. Vibrational spectroscopy combined with machine learning was applied to address this issue. The logistic regression model had the best F1 score with the highest %accuracy (73%), %sensitivity (73%), and %specificity (71%), followed by random forest and support vector machine models. Although the logistic regression model contributed significant findings, this study represents a laboratory research project. Because of the peculiarities of the ATR–FTIR spectral measurements, the spectra measured for several batches may differ, necessitating running the model on multiple spectral ranges and using increased sample sizes in subsequent applications. This proposed method has the potential to provide rapid and accurate results and may be valuable in future applications regarding toxin detection in foods when simple onsite testing is required.

Abstract Image

Creative Commons License

Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.