Table 9.
Comparison of the 8 classifiers, for the different experiments with the 100-miRNA signature
| TT vs | TCGA | GEO | ||||
|---|---|---|---|---|---|---|
| Classifier | TCGA | NT | GEO | (Subtype) | (Subtype) | Global |
| Gradient Boosting | 0.9359 | 0.9846 | 0.6697 | 0.9725 | 0.8909 | 0.8907 |
| Random Forest | 0.9324 | 0.9839 | 0.8085 | 0.9725 | 0.8634 | 0.9121 |
| Logistic Regression | 0.9237 | 0.9799 | 0.9351 | 0.9647 | 0.8476 | 0.9302 |
| Passive Aggressive | 0.8831 | 0.9606 | 0.8678 | 0.9556 | 0.8197 | 0.8974 |
| SGD | 0.9035 | 0.9767 | 0.9393 | 0.9490 | 0.8145 | 0.9166 |
| SVC | 0.9154 | 0.9791 | 0.7724 | 0.9451 | 0.8355 | 0.8895 |
| Ridge | 0.8305 | 0.9470 | 0.8867 | 0.9503 | 0.8300 | 0.8889 |
| Bagging | 0.9110 | 0.9812 | 0.7682 | 0.9555 | 0.9070 | 0.9046 |
Logistic Regression was the best across all experiments, and Ridge has the worst accuracy