| Type | avg. assigns. | avg. S
p
| avg. S
n
| Discriminant ability |
---|
OFDEG+GC (CP = 55%) | semi-supervised | 63.65% | 0.9400 | 0.9950 | 0.9675 |
OFDEG+GC | unsupervised | 97.33% | 0.9513 | 0.9525 | 0.9519 |
TF (CP = 75%) | semi-supervised | 83.44% | 0.9925 | 0.8925 | 0.9425 |
OFDEG+GC (CP = 75%) | semi-supervised | 77.75% | 0.8000 | 0.9625 | 0.8813 |
TF (CP = 55%) | semi-supervised | 69.28% | 1.0000 | 0.7450 | 0.8725 |
OFDEG | unsupervised | 97.34% | 0.9100 | 0.8300 | 0.8700 |
TF | unsupervised | 97.34% | 0.9905 | 0.6565 | 0.8235 |
- Discriminant ability is given by the average of the sensitivity (S
n
) and specificity (S
p
) values. In this case, we take the average of the average specificity and sensitivity over all tests conducted. We see that both the unsupervised and semi-supervised methods which use OFDEG+GC as a feature space perform best overall with respect to the simMC tests. Though the semi-supervised method outperforms the unsupervised method, the average number of assignments made by the unsupervised variant is far greater.