Research Article
A Comparison Study on Rule Extraction from Neural Network Ensembles, Boosted Shallow Trees, and SVMs
Table 14
Average predictive accuracy and average fidelity of rulesets by aggregating five models.
| Dataset | Avg. rules Pred. Acc. | Min rules Pred. Acc. | Max. rules Pred. Acc. | Avg. Fid. |
| Aust. Cred. Appr. | 86.7 (0.3) | 85.8 | 87.5 | 99.9 (0.1) | Breast Cancer | 96.8 (0.3) | 95.8 | 97.4 | 100.0 (0.0) | Breast Cancer 2 | 96.9 (0.4) | 95.1 | 97.7 | 99.9 (0.0) | Bupa Liv. Dis. | 72.2 (0.8) | 69.4 | 74.2 | 99.8 (0.1) | German Credit | 75.6 (1.1) | 72.7 | 76.9 | 99.9 (0.1) | Glass (binary) | 85.8 (2.1) | 77.8 | 89.0 | 99.8 (0.2) | Ionosphere | 93.3 (0.5) | 91.6 | 94.2 | 99.9 (0.1) | Musk1 | 90.2 (1.3) | 85.8 | 92.9 | 99.9 (0.1) | Promoters | 88.9 (1.6) | 83.0 | 91.7 | 99.6 (0.3) | Sonar | 86.8 (1.2) | 82.3 | 88.6 | 99.7 (0.1) |
|
|