Table 5.
Leave-one-out classification scores over the three data sets showing how accuracies and ADR F-scores are affected as one feature is removed from the set.
| Features | TW | DS | ADE | |||
|---|---|---|---|---|---|---|
| Accuracy | ADR F-score | Accuracy | ADR F-score | Accuracy | ADR F-score | |
| All | 86.2 | 0.538 | 83.6 | 0.678 | 88.2 | 0.812 |
| N-grams | 80.7 | 0.424 | 82.6 | 0.654 | 85.9 | 0.775 |
| UMLS STs and CUIs | 85.7 | 0.505 | 82.8 | 0.652 | 81.9 | 0.711 |
| Syn-set Expansions | 86.1 | 0.545 | 84.0 | 0.669 | 87.9 | 0.778 |
| Change Phrases | 87.1 | 0.521 | 83.9 | 0.665 | 88.0 | 0.803 |
| ADR Lexicon Match | 86.1 | 0.492 | 83.5 | 0.663 | 86.1 | 0.780 |
| Sentiword Score | 86.2 | 0.530 | 82.8 | 0.659 | 88.3 | 0.805 |
| Topics | 86.1 | 0.535 | 83.7 | 0.670 | 87.6 | 0.801 |
| Other Features | 86.9 | 0.534 | 83.6 | 0.677 | 88.1 | 0.809 |