Table 5 Accuracies for each individual class (bicarbonate, formaldehyde, peroxide, raw, starch, and sucrose) for multiclass classifications considering RF, GBM and CNN classifiers, in each of the selected training and test datasets: 90/10%, 75/25%, and 50/50%
From: On the utilization of deep and ensemble learning to detect milk adulteration
Classifier | Dataset | Bicarbonate | Formaldehyde | Peroxide | Raw | Starch | Sucrose |
RF | 90/10% | 0.7804 | 0.7209 | 0.7916 | 0.9883 | 0.8048 | 0.9636 |
75/25% | 0.7540 | 0.6884 | 0.7524 | 0.9839 | 0.8099 | 0.8773 | |
50/50% | 0.7634 | 0.6259 | 0.7847 | 0.9805 | 0.7444 | 0.8953 | |
GBM | 90/10% | 0.7804 | 0.7441 | 0.7291 | 0.9766 | 0.7560 | 0.9272 |
75/25% | 0.8032 | 0.6739 | 0.7623 | 0.9759 | 0.7768 | 0.8867 | |
50/50% | 0.7551 | 0.6259 | 0.7309 | 0.9780 | 0.7356 | 0.8870 | |
CNN | 90/10% | 0.9756 | 0.9302 | 0.8958 | 0.9844 | 0.9024 | 0.9636 |
75/25% | 0.9918 | 0.9057 | 0.9108 | 0.9887 | 0.9421 | 1.0000 | |
50/50% | 0.9958 | 0.9236 | 0.8340 | 0.9861 | 0.8854 | 0.9539 |