Table A.8.
Precision [P.], recall [R.], F1 score [F1] using UMLS ontology-guided feature extraction and mutual information.
| % of feature space | Total # of features | Naïve Bayes
|
SVM (linear kernel)
|
SVM (RBF kernel)
|
MaxEnt
|
||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| P. | R. | F1 | P. | R. | F1 | P. | R. | F1 | P. | R. | F1 | ||
|
| |||||||||||||
| 75.00 | 24,136 | 0.879 | 0.693 | 0.771 | 0.752 | 0.791 | 0.768 | 0.747 | 0.744 | 0.738 | 0.720 | 0.786 | 0.749 |
| 50.00 | 16,054 | 0.858 | 0.693 | 0.763 | 0.753 | 0.791 | 0.768 | 0.746 | 0.740 | 0.736 | 0.721 | 0.790 | 0.751 |
| 25.00 | 8,027 | 0.868 | 0.605 | 0.710 | 0.746 | 0.787 | 0.763 | 0.749 | 0.748 | 0.742 | 0.723 | 0.790 | 0.752 |
| 10.00 | 3,210 | 0.769 | 0.531 | 0.620 | 0.738 | 0.771 | 0.751 | 0.729 | 0.763 | 0.738 | 0.708 | 0.775 | 0.737 |
| 5.00 | 1,605 | 0.713 | 0.515 | 0.590 | 0.747 | 0.798 | 0.769 | 0.715 | 0.795 | 0.747 | 0.693 | 0.787 | 0.736 |
| 1.00 | 321 | 0.670 | 0.612 | 0.635 | 0.703 | 0.783 | 0.740 | 0.688 | 0.794 | 0.735 | 0.665 | 0.767 | 0.712 |
| 0.50 | 160 | 0.671 | 0.748 | 0.704 | 0.691 | 0.806 | 0.743 | 0.665 | 0.821 | 0.731 | 0.660 | 0.810 | 0.726 |
| 0.10 | 32 | 0.613 | 0.833 | 0.705 | 0.685 | 0.771 | 0.725 | 0.646 | 0.791 | 0.707 | 0.643 | 0.810 | 0.715 |
| 0.05 | 16 | 0.575 | 0.849 | 0.685 | 0.650 | 0.767 | 0.702 | 0.614 | 0.845 | 0.707 | 0.614 | 0.810 | 0.698 |
Note: Performance of shaded region is not significantly different from the top overall model (i.e., F1 of 77.4%).