Table 2.
Top ten best performing models using all sequence and sequence-specific features sets ordered by ROC AUC.
| All sequences | Sequence-specific | |||||
|---|---|---|---|---|---|---|
| Model | Feature reduction | ROC AUC mean (SD) | Model | Feature reduction | Sequence | ROC AUC mean (SD) |
| Lasso | Full | 0.953 (0.041) | Lasso | Full | F | 0.951 (0.049) |
| Enet | Full | 0.952 (0.038) | Enet | Full | F | 0.948 (0.049) |
| GBRM | Full | 0.940 (0.040) | RF | Full | F | 0.947 (0.042) |
| RF | Full | 0.937 (0.046) | ada | Full | F | 0.945 (0.038) |
| ada | Full | 0.933 (0.045) | Lasso | Full | CE | 0.943 (0.041) |
| Ridge | Full | 0.928 (0.053) | GBRM | Full | F | 0.943 (0.045) |
| GBRM | Corr | 0.925 (0.056) | GBRM | Lincomb | F | 0.943 (0.045) |
| SVMRad | Full | 0.921 (0.042) | Lasso | Full | T1 | 0.941 (0.046) |
| RF | Corr | 0.920 (0.045) | Enet | Full | T1 | 0.941 (0.041) |
| MLP | Full | 0.919 (0.055) | GBRM | Corr | F | 0.940 (0.046) |
Enet: elastic net; Lasso (least absolute shrinkage and selection operator; GBRM: generalized boosted regression model; RF: random forest; ada: boosting of classification trees with adaBoost; SVMRad: SVM with a radial kernel; MLP: multi-layer perceptron; full: full feature set; corr: High correlation filter; lincomb: linear combinations filter; F: FLAIR; CE: T1-CE.