Table 1.
Summary of reviewed mycotoxin detection studies. In cases where more than one ML model is used, the highest-performing model is reported.
| Study | Data Type | ML Model | Application Context | Accuracy |
|---|---|---|---|---|
| Camardo et al., 2021 [38] | Spatiotemporal | NN | Corn in Northern Italy (2005–2018) | >75% |
| Niedbala et al., 2020 [57] | Spatiotemporal | NN | Winter wheat in Poland (2011–2013) | DONANN: 99%, NIVANN: 81% |
| Jubair et al., 2021 [59] | Spatiotemporal | GPTransformer | Barley in Canada (2014–2015) | Not significantly better than BLUP |
| Rangarajan et al., 2022 [43] | Hyperspectral | CNN (DarkNet 19) | Wheat for Fusarium head blight | 100% |
| Qiu et al., 2019 [63] | Hyperspectral | CNN | Wheat for Fusarium head blight | 92% |
| Jin et al., 2018 [62] | Hyperspectral | CNN (2D conv. bidirectional GRU) | Wheat for Fusarium head blight | 84.6% |
| Han et al., 2019 [49] | Hyperspectral | CNN | Peanuts for aflatoxin | 95% |
| Han et al., 2019 [49] | Hyperspectral | CNN | Corn and peanuts for aflatoxin | 96% corn, 92% peanuts |
| Gao et al., 2021 [65] | Hyperspectral | 1D-CNN | Peanuts and corn for aflatoxin | Peanuts: 96.4%, corn: 92.1% |
| Öner et al., 2019 [66] | Infrared spectroscopy | MLP NN, RF, SVM, adaptive boosting | Corn for fungal contamination | MLP: 91% |
| Leggieri et al., 2021 [71] | E-nose | NN, LR, DA | Corn for aflatoxin and fumonisins | NN: 78% (aflatoxin), 77% (fumonisins) |
| Ghilardelli et al., 2022 [75] | NIR | RF | Corn silage | >90% |
| Teixidó et al., 2023 [76] | NIR | RF | Oat for DON | 77.8% |
| Ma et al., 2023 [78] | Biosensor | sPLS-DA, svmLinear, svmRadial, RF, NN, HDDA | Peanuts and corn | RF: 95–98% |
| Tarazona et al., 2021 [79] | Spatiotemporal | NN, RF, XGB, MLR | Corn | RF: Best performance |
| Mateo et al., 2023 [80] | Spatiotemporal | NN, RF, XGB, MLR | Cereals for lactic acid bacteria | RF, XGB: Similar performance |
| Chávez et al., 2022 [92] | NIR | GBM, RF, LASSO, GLMNET, SVM | Single kernel corn | GBM: 83% |
| Liu et al., 2018 [103] | Spatiotemporal | BN, LR, mechanistic | Wheat for DON | LR: 88%, BN: 86% |
| Guo et al., 2020 [104] | TRNs | BN | Fusarium graminearum | High confidence modules: 81.8% |
| Kim et al., 2024 [33] | Weather | CNN | Wheat grains stored in sealed containers | 83.3% |
| Castano et al., 2023 [34] | Weather | NN | Corn in the US | Aflatoxin: 73%, fumonisin: 85% |
| Branstad et al., 2023 [39] | Spatiotemporal | GBM | Corn in Iowa | 20 ppb: 96.8%, 5 ppb: 90.3% |
| Wang et al., 2022 [87] | Spatiotemporal | XGB | Peanuts and soybeans in China, Brazil, Argentina | >90% |
| Xie et al., 2022 [88] | Spatiotemporal | XGB | Peanuts in China | 98% |
| Xie et al., 2022 [88] | Spatiotemporal | XGB with decision rule | Peanuts in China | 87.2% |
| Zhao et al., 2017 [97] | Hyperspectral | SVM | Corn in China | 91.67% |
| Kim et al., 2023 [96] | Hyperspectral | SVM | Ground corn samples from Texas | 95.7% |
| Almoujahed et al., 2022 [45] | Spectral | SVM | Fusarium head blight in wheat in Belgium, 2020–2021 | 97% |
| Kos et al., 2016 [99] | Spatiotemporal | CART | Wheat in Italy | 88–92% |
| Purchase et al., 2023 [100] | Spatiotemporal | DT | Breakfast cereals in Italy | High fumonisin risk in high sodium or high-fat cereals |