Course contentsShow
← Machine Learning and Deep Learning
Lesson 3284 of 3,53872. Bias and FairnessFree lesson

Equalized Odds

Understand equalized odds: requiring both equal true positive rates and equal false positive rates across groups simultaneously.

Equalized Odds

What You'll Learn

You'll learn how equalized odds extends equal opportunity by requiring fair performance both ways—for positive and negative outcomes across all protected groups.


The Core Idea

Equal opportunity (the previous concept) only required equal true positive rates across groups—meaning the model should be equally good at identifying positives for everyone. But that's only half the fairness story.

Equalized odds is more demanding: it requires both equal true positive rates and equal false positive rates across groups simultaneously.

Think of it like this: A loan approval model satisfies equalized odds if:

  1. Among qualified applicants, approval rates are equal across race/gender (equal TPR)
  2. Among unqualified applicants, rejection rates are equal across race/gender (equal FPR)

This means the model makes both types of correct decisions fairly—it doesn't favor one group with more lenient standards (higher FPR) while being stricter with another.


Mathematical Constraint

For protected groups A and B:

TPR(A) = TPR(B) and FPR(A) = FPR(B)

Or equivalently:

  • P(Ŷ=1 | Y=1, A) = P(Ŷ=1 | Y=1, B)
  • P(Ŷ=1 | Y=0, A) = P(Ŷ=1 | Y=0, B)

Why It Matters

Equalized odds prevents subtle biases where a model might correctly identify positives equally (satisfying equal opportunity) but make wildly different error rates. For instance, falsely flagging innocent individuals at different rates across racial groups in criminal justice violates equalized odds even if true conviction rates match.

It's a stronger fairness guarantee than equal opportunity alone, ensuring balanced performance across the entire confusion matrix.


Key Takeaway: Equalized odds requires equal true positive rates and equal false positive rates across protected groups, ensuring both correct acceptance and correct rejection happen fairly for everyone.