Summary of De-biasing Models Of Biased Decisions: a Comparison Of Methods Using Mortgage Application Data, by Nicholas Tenev
De-Biasing Models of Biased Decisions: A Comparison of Methods Using Mortgage Application Data
by Nicholas Tenev
First submitted to arxiv on: 1 May 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Computers and Society (cs.CY); Econometrics (econ.EM)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This paper explores how machine learning models like XGBoost can inherit biases from the data they’re trained on, even when ethnicity isn’t used as a predictive variable. By adding counterfactual ethnic bias to real mortgage application decisions, researchers show that this bias is replicated by the model. To address this issue, several de-biasing methods are compared: averaging over prohibited variables, taking the most favorable prediction over prohibited variables (a novel approach), and jointly minimizing errors as well as association between predictions and prohibited variables. While de-biasing can recover some original decisions, results depend on whether bias is introduced through a proxy. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Machine learning models can help make decisions like loan approvals more efficient, but they might also pick up biases from the data they’re trained on. Researchers made real mortgage application data a little worse by adding fake biases against certain groups. They then trained a machine learning model (XGBoost) to see if it would still be biased. Even though ethnicity wasn’t used as a predictor, the model picked up the bias. To fix this, different ways of “debiasing” were tested: taking averages, choosing the best prediction, and trying to balance errors with fairness. It’s good news that some of the original decisions can be recovered, but it depends on how the bias was introduced. |
Keywords
» Artificial intelligence » Machine learning » Xgboost
