Research Brief

IFEM: Intersectional Fairness Evaluation

IFEM is an intersectional fairness evaluation framework designed to expose subgroup-level disparities that may be hidden by aggregate accuracy or single-attribute fairness analysis.

Why This Matters

High overall performance can conceal unequal behaviour across combined demographic groups. In face recognition and other sensitive AI systems, intersectional subgroup analysis can reveal reliability differences that may not be visible when race, gender, or age are evaluated separately.

IFEM supports AI assurance by examining reliability gaps across intersectional groups that may be obscured by aggregate evaluation. This provides additional evidence for deployment review where aggregate performance alone may not capture relevant variation in fairness and subgroup reliability.

Key Contributions

• Introduces an intersectional framework for subgroup fairness evaluation.
• Evaluates disparities beyond aggregate accuracy and single-attribute analysis.
• Supports threshold-sensitive fairness assessment in face recognition systems.
• Examines how mitigation strategies affect subgroup reliability and fairness gaps.

Deployment Relevance

IFEM supports AI assurance by identifying hidden reliability gaps across intersectional groups. This strengthens deployment evaluation where aggregate performance alone may give a misleading picture of operational fairness and subgroup reliability.