Table 5.
Overall performance (average precision) changes for different dataset, feature and classifier combinations
| Training set | Triage |
CTD |
||
|---|---|---|---|---|
| Feature | Multiword features |
All proposed features |
||
| Classifier | Bayes | Huber | Huber | Huber |
| 2-Acetylaminofluorene | 0.7151 | 0.6812 | 0.7055 | 0.6932 |
| Amsacrine | 0.5880 | 0.6676 | 0.6850 | 0.7411 |
| Aniline | 0.7589 | 0.7646 | 0.8000 | 0.7708 |
| Aspartame | 0.3755 | 0.4520 | 0.4890 | 0.5902 |
| Doxorubicin | 0.8434 | 0.8718 | 0.8689 | 0.8895 |
| Indomethacin | 0.9599 | 0.9699 | 0.9761 | 0.9626 |
| Quercetin | 0.9068 | 0.9176 | 0.9321 | 0.9227 |
| Raloxifene | 0.7913 | 0.7940 | 0.8175 | 0.7759 |
| Average performance | 0.7424 | 0.7648 | 0.7843 | 0.7933 |
‘Triage’ means the Triage training set is used for training. ‘CTD’ means the full CTD set is used to augment the positive set and negatives are from the Triage set. Again a leave-one-out train and test scenario are used. ‘Bayes’ and ‘Huber’ indicate Bayes and Huber classifiers, respectively.