Definition
Measures the degree to which AI-generated adverse action reasons accurately reflect the features that actually drove the model's denial decision, distinguishing genuine explanations from post-hoc approximations.
Formula
AAEF = (1/S) x SUM |Reasons_stated ∩ Features_actual| / N
Proposed formula, pending validation pilots.
At a Glance
Category
AI Credit & Lending Decision Quality
Decision Supported
Are the adverse action notices sent to denied applicants legally defensible explanations of the actual decision logic?