Table 3:
Model | Sensitive attribute: Race | Sensitive attribute: Gender | ||||
ROC AUC | DPD | EOD | ROC AUC | DPD | EOD | |
MELD-score | 0.505±0.000 | — | — | 0.505±0.000 | — | — |
Logistic Regression | 0.777±0.000 | 0.648±0.017 | 0.834±0.007 | 0.777±0.000 | 0.021±0.000 | 0.033±0.001 |
Random forest | 0.804±0.000 | 0.630±0.030 | 0.703±0.047 | 0.804±0.000 | 0.020±0.001 | 0.036±0.001 |
GBDT | 0.809±0.000 | 0.637±0.027 | 0.713±0.033 | 0.809±0.000 | 0.017±0.000 | 0.031±0.001 |
W/o first-step | 0.793±0.000 | 0.596±0.022 | 0.687±0.038 | 0.792±0.000 | 0.016±0.002 | 0.027±0.002 |
W/o second-step | 0.793±0.001 | 0.616±0.041 | 0.745±0.076 | 0.793±0.001 | 0.014±0.007 | 0.026±0.009 |
Ours | 0.792±0.000 | 0.597±0.015 | 0.662±0.029 | 0.793±0.001 | 0.011±0.001 | 0.022±0.003 |