We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.LG

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Machine Learning

Title: Characterizing Fairness Over the Set of Good Models Under Selective Labels

Abstract: Algorithmic risk assessments are used to inform decisions in a wide variety of high-stakes settings. Often multiple predictive models deliver similar overall performance but differ markedly in their predictions for individual cases, an empirical phenomenon known as the "Rashomon Effect." These models may have different properties over various groups, and therefore have different predictive fairness properties. We develop a framework for characterizing predictive fairness properties over the set of models that deliver similar overall performance, or "the set of good models." Our framework addresses the empirically relevant challenge of selectively labelled data in the setting where the selection decision and outcome are unconfounded given the observed data features. Our framework can be used to 1) replace an existing model with one that has better fairness properties; or 2) audit for predictive bias. We illustrate these uses cases on a real-world credit-scoring task and a recidivism prediction task.
Comments: Added comparison methods to the empirical lending analysis
Subjects: Machine Learning (cs.LG); Machine Learning (stat.ML)
Cite as: arXiv:2101.00352 [cs.LG]
  (or arXiv:2101.00352v3 [cs.LG] for this version)

Submission history

From: Amanda Coston [view email]
[v1] Sat, 2 Jan 2021 02:11:37 GMT (1838kb,D)
[v2] Wed, 13 Jan 2021 00:25:36 GMT (1838kb,D)
[v3] Sat, 1 May 2021 00:21:59 GMT (3185kb,D)

Link back to: arXiv, form interface, contact.