Evaluating the Fairness of Predictive Student Models Through Slicing Analysis
summary
Proposes a concrete slicing procedure that measures fairness as the integral distance between subgroup ROC curves, named ABROCA, and applies it to several prior predictive models evaluated over 45 MOOCs (replicating some earlier work). Finds the area-between-curves reveals threshold ranges where predictions diverge by subgroup, reports a clear link between gender balance skew and larger ABROCA, and shows no AUC–fairness trade-off in general, implying fairness can be improved without losing overall classifier performance.
AI-written summary · about
chapters (6)
- 0:00Introduction and motivation
- 2:52Fairness in predictive models
- 6:02Slicing analysis methodology
- 10:35ABROCA fairness metric
- 12:42Experimental results and findings
- 14:50Q&A and conclusion
related talks
28mMeasuring Model Fairness
30mBuilding fairer models for finance
42mFairness in Machine Learning: From Theory to Practice
44mProcedural Justice in Algorithmic Fairness: Transparency, Control and Participation Perspectives
35mOpen Source Fairness
31mMeasuring Model Fairness
34mInterpretability and Algorithmic Fairness
39mBeyond Bias: Crafting Fairness in Pythonic Machine Learning Systems