Score imbalance, prediction gaps, and mitigation readiness. Review proxy effects, governance strength, and fairness drift. Make model risk decisions using transparent weighted evidence today.
Enter measured gaps as percentages. Enter controls as adequacy percentages, where higher adequacy lowers risk.
Direct Gap Risks
Representation, false positive, false negative, calibration, label bias, and proxy dependence risks use their entered percentage directly.
Selection Parity Risk
Selection Parity Risk = (1 − Selection Rate Ratio) × 100
Control Risks
Control Risk = 100 − Adequacy Score
Context Risks
Context Risk = ((Score − 1) / 4) × 100 for regulatory sensitivity and impact severity.
Overall Weighted Bias Score
Overall Score = Σ (Component Risk × Weight)
Weighting emphasizes measurable fairness gaps, data quality, feature proxy risk, governance maturity, and contextual exposure. The final score is reported on a 0 to 100 scale.
| Metric | Example Input | Normalized Risk | Comment |
|---|---|---|---|
| Representation Gap | 18% | 18.00 | Moderate representation difference across groups. |
| Selection Rate Ratio | 0.82 | 18.00 | Parity is weakened because ratio is below 1.00. |
| False Positive Rate Gap | 12% | 12.00 | Some groups receive more false alarms. |
| False Negative Rate Gap | 15% | 15.00 | Missed outcomes differ across groups. |
| Label Bias Risk | 20% | 20.00 | Annotation practices may amplify imbalance. |
| Proxy Feature Dependence | 30% | 30.00 | Proxy variables need stronger review. |
| Oversight Adequacy | 70% | 30.00 | Human review exists, but is not fully strong. |
| Documentation Completeness | 65% | 35.00 | Documentation is usable but incomplete. |
| Mitigation Maturity | 60% | 40.00 | Bias controls exist with room to improve. |
| Monitoring Coverage | 55% | 45.00 | Post-release surveillance is weaker than preferred. |
| Regulatory Sensitivity | 4 | 75.00 | Use case attracts stronger oversight pressure. |
| Impact Severity | 5 | 100.00 | Potential harm from unfair outcomes is very high. |
| Overall Score | Weighted result | 26.26 | Guarded risk with stronger context concern. |
It estimates overall bias exposure in an AI system by combining data imbalance, fairness gaps, governance controls, and contextual severity into one weighted score.
No. A low score suggests lower visible risk, but fairness still depends on test design, subgroup coverage, changing data, and ongoing real-world monitoring.
That format expresses parity directly. A value of 1 means equal selection rates, while lower values indicate stronger disparity and therefore higher selection parity risk.
They are entered as strengths. Stronger governance reduces exposure, so the calculator converts adequacy into risk by subtracting the adequacy score from 100.
Yes. It is especially useful before launch, during model review, or before procurement approval when teams need a structured bias-risk snapshot.
Use the best validated figures available. For stronger decisions, compare training, validation, and live monitoring results rather than relying on one dataset alone.
Many teams begin formal remediation once scores enter the Significant range or when a single driver is extremely high, even if the overall score appears moderate.
Yes. You can edit the weights in the code to match your governance framework, sector rules, or internal materiality thresholds.
Important Note: All the Calculators listed in this site are for educational purpose only and we do not guarentee the accuracy of results. Please do consult with other sources as well.