Table 3 Group Fairness Metrics

From: Improving medical machine learning models with generative balancing for equity and excellence

 

eICU

MIMIC-IV

Training Data

Age

Gender

Race

Insurance

 

Disparate Impact

Theil Index

Disparate Impact

Theil Index

Disparate Impact

Theil Index

Disparate Impact

Theil Index

Original Data

0.772

0.0137

0.971

0.1566

1.217

0.0106

0.907

0.0015

Upsampling

0.898

0.0058

0.962

0.0760

1.165

0.0119

0.946

0.0004

Downsampling

0.961

0.0023

0.992

0.0419

1.203

0.0112

0.922

0.0007

Separate Models

0.917

0.1348

1.024

0.3003

1.083

0.0342

0.966

0.0033

SMOTE

0.912

0.0097

0.978

0.1233

1.108

0.0260

1.008

0.0008

MCRAGE

0.757

0.0142

0.968

0.2100

1.232

0.0104

0.896

0.0016

SMA

0.878

0.0093

0.992

0.0606

–

–

–

–

FairPlay

1.038

0.0026

0.991

0.0519

1.197

0.0105

0.923

0.0011

  1. Bold fonts indicates the best performing model.
Back to article page