Unit 10: Bias Detection, Testing & Mitigation for Devs
Scenario: A fintech company deployed a credit scoring model.
Issue: Lower approval rates for women and ethnic minorities.
What Went Wrong:
Training data reflected historical approval biases
Model penalized applicants with short credit histories (disproportionately affecting marginalized groups)
✅ What Fixed It:
Applied reweighting + feature removal
Conducted fairness audit + retrained
Updated system docs + dashboarded impact
