Unit 10: Bias Detection, Testing & Mitigation for Devs
Bias testing = checking whether the model treats groups differently.
📊 Top Metrics:
Demographic Parity – Are outcomes evenly distributed?
Equal Opportunity – Are success rates consistent?
Predictive Parity – Are precision/recall balanced?Disparate Impact – Legal red flag if ratio < 0.8
🧪 Tools of the Trade:
IBM AI Fairness 360
Fairlearn (Python)
What-If Tool (Google)
Amazon SageMaker Clarify
