Algorithmic Fairness Audit
Real-time fairness metrics for all ML prediction models. Threshold: Disparate Impact Ratio ≥ 0.80 (four-fifths rule).
2
Models Passed
2
Models Warning
0
Models Failed
0.80
DI Threshold
Gambling Risk
v1.0.0 | XGBoostDisparate Impact Ratio
0.87
Demographic Parity
warningEqualized Odds
passedCalibration
0.94
0.00.80 threshold1.0
Recommendations
- - Review elder subgroup predictions — calibration slightly lower
- - Monitor linguistically isolated subgroup for potential bias
Dropout Risk
v1.0.0 | Random ForestDisparate Impact Ratio
0.92
Demographic Parity
passedEqualized Odds
passedCalibration
0.96
0.00.80 threshold1.0
Recommendations
- - Model passes fairness checks. Continue monitoring.
Mental Health Risk
v1.0.0 | Gradient Boosted TreesDisparate Impact Ratio
0.84
Demographic Parity
warningEqualized Odds
passedCalibration
0.91
0.00.80 threshold1.0
Recommendations
- - Gender subgroup shows differential calibration
- - Consider adding fairness constraints to next training round
Elder Isolation Risk
v1.0.0 | Logistic RegressionDisparate Impact Ratio
0.95
Demographic Parity
passedEqualized Odds
passedCalibration
0.93
0.00.80 threshold1.0
Recommendations
- - Interpretable model with strong fairness metrics.