Table 3.
Application of two different algorithms (linear support vector machine and ensemble bagged decision tree) to the five(5) unique SFRP data sets; [100][D1], [100][D2], [100][D3], [100][D4] and [100][D5]
| Class | Frequency (%) | ||||
|---|---|---|---|---|---|
| D1 | D2 | D3 | D4 | D5 | |
| 0 | 56 | 43 | 37 | 43 | 53 |
| 1 | 14 | 20 | 27 | 18 | 17 |
| 2 | 5 | 18 | 7 | 15 | 13 |
| 3 | 19 | 17 | 11 | 24 | 17 |
| 4 | 6 | 2 | 18 | 0 | 0 |
| SVM-L accuracy (%) | 78.3 ± 0.5 | 92.7 ± 0.5 | 78.3 ± 2.4 | 88.3 ± 0.5 | 86.7 ± 0.9 |
| EBDT accuracy (%) | 83.3 ± 1.2 | 96.3 ± 0.9 | 72.3 ± 0.9 | 90.0 ± 0.0 | 87.7 ± 1.2 |
| Class with the highest confusion (TPR—sensitivity for EBDT) | 4 (17%) | 4 (0%) | 2 (14%) | 2 (60%) | 2 (77%) |
The table presents the fraction of the total observations for each class for each dataset and corresponding cross-validated accuracies for both classifiers. The confusion (true positive rates (TPR)) was correlated to the percentage of observations of the class