Table 2 Cross-validation with the blind-tested samples and the respective accuracy, sensitivity, specificity, precision, negative predictive value, false alarm, and F1 score. Comparison of drift effect by excluding and including sugar (glucose, fructose, and sucrose) adulterants from the same batch of blind-tested samples in the training dataset. A comparison analysis by including and excluding other adulterants not excluded from the training dataset was also performed. Hydrogen peroxide (H2O2), sodium hydroxide (NaOH).
From: Non-targeted detection of food adulteration using an ensemble machine-learning model
| Â | Testing of adulterant(s) blinded to training | |||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Does not consider drift effect | Considers drift effect | |||||||||||||||||||
Training with sugar adulterants, testing of normal raw milk samples from the cross-validation batch | Training with sugar and other adulterants, testing of normal raw milk samples from the cross-validation batch | |||||||||||||||||||
H2O2 | NaOH | salt | H2O2, NaOH | H2O2, salt | NaOH, salt | H2O2, NaOH, salt | H2O2 | NaOH | salt | H2O2, NaOH | H2O2, salt | NaOH, salt | H2O2, NaOH, salt | H2O2 | NaOH | salt | H2O2, NaOH | H2O2, salt | NaOH, salt | |
Accuracy | 0.5120 | 0.5280 | 0.5120 | 0.4714 | 0.4571 | 0.4714 | 0.4258 | Undetermined | 0.8625 | 0.8875 | 0.8316 | 0.7895 | 0.9053 | 0.8182 | 0.8125 | 0.8625 | 0.8500 | 0.8316 | 0.8211 | 0.8947 |
Sensitivity | 0.0500 | 0.0667 | 0.0500 | 0.0533 | 0.0400 | 0.0533 | 0.0444 | Undetermined | 1.0000 | 0.9333 | 0.7333 | 0.6000 | 0.9667 | 0.7333 | 0.8667 | 1.0000 | 1.0000 | 0.8667 | 0.7000 | 1.0000 |
Specificity | 0.9385 | 0.9538 | 0.9385 | 0.9538 | 0.9385 | 0.9538 | 0.9538 | Undetermined | 0.8308 | 0.8769 | 0.8769 | 0.8769 | 0.8769 | 0.8769 | 0.8000 | 0.8308 | 0.8154 | 0.8154 | 0.8769 | 0.8462 |
Precision | 0.4286 | 0.5714 | 0.4286 | 0.5714 | 0.4286 | 0.5714 | 0.5714 | Undetermined | 0.5769 | 0.6364 | 0.7333 | 0.6923 | 0.7838 | 0.8049 | 0.5000 | 0.5769 | 0.5556 | 0.6842 | 0.7241 | 0.7500 |
Negative predictive value | 0.5169 | 0.5254 | 0.5169 | 0.4662 | 0.4586 | 0.4662 | 0.4189 | Undetermined | 1.0000 | 0.9828 | 0.8769 | 0.8261 | 0.9828 | 0.8261 | 0.9630 | 1.0000 | 1.0000 | 0.9298 | 0.8636 | 1.0000 |
False alarm | 0.0615 | 0.0462 | 0.0615 | 0.0462 | 0.0615 | 0.0462 | 0.0462 | Undetermined | 0.1692 | 0.1231 | 0.1231 | 0.1231 | 0.1231 | 0.1231 | 0.2000 | 0.1692 | 0.1846 | 0.1846 | 0.1231 | 0.1538 |
F1 score | 0.0896 | 0.1194 | 0.0896 | 0.0976 | 0.0732 | 0.0976 | 0.0825 | Undetermined | 0.7317 | 0.7568 | 0.7333 | 0.6429 | 0.8657 | 0.7674 | 0.6341 | 0.7317 | 0.7143 | 0.7647 | 0.7119 | 0.8571 |