From: Fair automated assessment of noncompliance in cargo ship networks
Measure | Baseline | Random forest | Fair random forest |
---|---|---|---|
precision (non-white) | 97.1% | 89.8% | 89.0% |
precision (white) | 95.2% | 87.7% | 86.1% |
recall (non-white) | 42.3% | 75.5% | 82.5% |
recall (white) | 5.2% | 88.6% | 86.6% |
F1 (non-white) | 58.9% | 82.0% | 85.6% |
F1 (white) | 9.9% | 88.2% | 86.4% |
\(\epsilon_{\text{parity}}\) | 0.317 | 0.099 | 0.023 |
\(\epsilon_{\text{odds}}\) | 0.371 | 0.132 | 0.040 |
AUCY | 0.543 ± 0.006 | 0.814 ± 0.004 | 0.776 ± 0.008 |
AUCS | 0.672 ± 0.010 | 0.627 ± 0.014 | 0.538 ± 0.011 |