simonpcouch / cascadia-24 Goto Github PK

This repository contains source code and slides for the talk "Fair machine learning" at Cascadia R Conf in June 2024. The slides for the talk are available here.

To learn more about machine learning with R:

• Machine learning with tidymodels: tmwr.org
• More example notebooks with tidymodels are at tidymodels.org. Two fairness-oriented ones:
  • Are GPT detectors fair?, focused on how different fairness metrics encoded different interpretations of fairness across linguistic proficiency.
  • Fair prediction of hospital readmission, focused on training models that are near-fair with respect to a set of fairness metrics across racial groups.
• An overview of our reading group's conclusions and the kinds of fairness-oriented workflows we decided to support.

I mentioned these works in the talk:

• ccao-data/public from folks at the Cook County Assessor's Office—source code for a talk that inspired the property assessment example.
• An Unfair Burden by Jason Grotto in 2017, an article from the Chicago Tribune investigating unfairness in Cook County's property tax system.
• How Lower-Income Americans Get Cheated on Property Taxes from the NYTimes Editorial Board in 2021, which performs a similar analysis across counties in the US and additionally engages with the racial disparities in property taxation.
• Algorithmic Fairness: Choices, Assumptions, and Definitions from Mitchell et al. in 2021. A great reference for the various choices and assumptions that underlie fairness-oriented analysis of ML models.

In this repository,

• index.qmd contains the source code for the slides. The slides use images in the /figures directory.
• /docs is auto-generated from index.qmd. Content in that folder is likely unhelpful for a human reader, and is better viewed at the links above. :)