Dictionary · Trustworthy Machine Learning
L2 — definitions grouped by regulatory framework.
3 senses under Trustworthy Machine Learning
Nouns
3 senses- Adverse Impact Ratio
privileged and unprivileged groups receiving different outcomes irrespective of the decision maker’s intent and irrespective of the decision-making procedure. Quantified as the ratio: disparate impact ratio = 𝑃( 𝑦̂ (𝑋) = fav ∣∣ 𝑍 = unpr )/𝑃( 𝑦̂ (𝑋) = fav ∣∣ 𝑍 = priv ) where 𝑃(𝑦̂ (𝑋) = fav) is the favorable label, (𝑍 = priv) is the privileged group, and (𝑍 = unpr) is the unprivileged group.
- False Positive
A false positive is approving an applicant who should be denied
- False Negative
A false negative is denying an applicant who should be approved