Table 1.
Confusion matrix of GBDT, ANN, and SVM classification for crisp grass carp. (a) GBDT. Accuracy of all samples = 92.00%; Accuracy of crisp grass carp samples = 91.67%. (b) ANN. Accuracy of all samples = 89.00%; Accuracy of crisp grass carp samples = 85.42%. (c) SVM. Accuracy of all samples = 91.00%; Accuracy of crisp grass carp samples = 87.50%.
| (a) GBDT model | ||||||||
|---|---|---|---|---|---|---|---|---|
| Target | Level 1 | Level 2 | Level 3 | Level 4 | Level 5 | Sensitivity | ASC | ASA |
| level 1 | 48 | 3 | 0 | 1 | 0 | 92.31% | - | 91.80% |
| level 2 | 0 | 12 | 0 | 0 | 0 | 100% | 92.00% | |
| level 3 | 3 | 1 | 8 | 0 | 0 | 66.67% | ||
| level 4 | 0 | 0 | 0 | 12 | 0 | 100% | ||
| level 5 | 0 | 0 | 0 | 0 | 12 | 100% | ||
| Specificity | 94.12% | 75.00% | 100% | 90.91% | 100% | - | - | - |
| APC | - | 91.485 | - | - | - | |||
| APA | 92.01% | - | - | - | ||||
| (b) ANN model | ||||||||
| Target | Level 1 | Level 2 | Level 3 | Level 4 | Level 5 | Sensitivity | ASC | ASA |
| level 1 | 48 | 3 | 0 | 1 | 0 | 92.31% | - | 86.79% |
| level 2 | 1 | 9 | 2 | 0 | 0 | 75.00% | 85.00% | |
| level 3 | 0 | 2 | 10 | 0 | 0 | 83.33% | ||
| level 4 | 0 | 0 | 0 | 10 | 2 | 83.33% | ||
| level 5 | 0 | 0 | 0 | 0 | 12 | 100% | ||
| Specificity | 97.95% | 64.26% | 83.33% | 90.91% | 85.71% | - | - | - |
| APC | - | 81.05% | - | - | - | |||
| APA | 84.43% | - | - | - | ||||
| (c) SVM model | ||||||||
| Target | Level 1 | Level 2 | Level 3 | Level 4 | Level 5 | Sensitivity | ASC | ASA |
| level 1 | 49 | 2 | 0 | 1 | 0 | 94.23% | - | 85.51% |
| level 2 | 0 | 8 | 3 | 0 | 0 | 66.67% | 83.00% | |
| level 3 | 0 | 3 | 8 | 0 | 0 | 66.67% | ||
| level 4 | 0 | 0 | 12 | 0 | 100% | |||
| level 5 | 0 | 0 | 0 | 0 | 12 | 100% | ||
| Specificity | 100% | 61.54% | 72.73% | 92.31% | 100% | - | - | - |
| APC | - | 81.65% | - | - | - | |||
| APA | 85.32% | - | - | - | ||||
Note: ASC—average sensitivity of crisp grass carp samples; ASA—average sensitivity of all testing samples; APC—average specificity of crisp grass carp samples; APA—average specificity of all testing samples. Accuracy is the proportion of properly predicted samples to the total sample. Sensitivity and specificity are calculated as described in the Materials and Method. Green color denotes correct classification, red color denotes false classification.