Building fairer models for finance
summary
Uses the Home Credit Default Risk dataset (≈300k rows) and a simple logistic regression to show fairness wins without wrecking accuracy: baseline model AUC ≈ 0.67 with true positive rate ratio ~64% and false positive rate ratio ~50% versus the majority. Removing gender barely lowers AUC while boosting TPR ratio to ~79% and FPR to ~70%; training with gender but omitting it at serving pushes TPR ≈91% and FPR ≈88%. Discusses 80% rule, reject inference, and LIME/SHAP for explanations.
AI-written summary · about
chapters (13)
- 0:00Fairer finance models
- 1:07Fairness in US financial regulation
- 4:43Fairness in finance and law
- 8:14UK fairness approach
- 10:01Risk models with fairness
- 12:17Demo: gender bias in lending
- 15:29Quantifying fairness metrics
- 18:22Mitigating bias without performance loss
- 22:00Fairness metrics by use case
- 24:11Detecting bias without protected data
- 25:33Future of ML in lending
- 28:01Model fairness process
- 29:47Conclusion
related talks
01h 30mFairness Tales: How To Measure And Mitigate Unfair Bias in Machine Learning Models
39mBeyond Bias: Crafting Fairness in Pythonic Machine Learning Systems
48mResponsible AI
14mBuilding with the Responsible AI toolkit
28mBeyond Prediction Performance: How Modeling Decisions Shape Fairness Outcomes in Statistical...
31mMeasuring Model Fairness
41mLow rate loans for ladies, stags pay extra: The Role of Ethics in AI/ML
01h 31mHow To Measure And Mitigate Unfair Bias in Machine Learning Models