Summary of The Impact Of Differential Feature Under-reporting on Algorithmic Fairness, by Nil-jana Akpinar et al.
The Impact of Differential Feature Under-reporting on Algorithmic Fairness
by Nil-Jana Akpinar, Zachary C. Lipton, Alexandra Chouldechova
First submitted to arxiv on: 16 Jan 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Computers and Society (cs.CY); Machine Learning (stat.ML)
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 Predictive risk models in the public sector are developed using administrative data, which is more complete for subpopulations relying on public services. This leads to biases in algorithmic decision-making, with differential feature under-reporting being a significant driver of disparities. While prior work has examined additive feature noise and missing features, this setting remains understudied. We present an analytically tractable model of differential feature under-reporting and demonstrate its impact on algorithmic fairness. Standard methods fail to mitigate bias, so we propose new methods tailored to this setting. Our results show that under-reporting leads to increasing disparities in real-world data settings, but our proposed solutions are successful in mitigating unfairness. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Predictive risk models used by government agencies can be biased because they only have complete information about people who rely heavily on public services. This is a problem because it means the algorithms might not work fairly for everyone. The problem is that some data points are missing, but this is not always marked as “missing”. We studied how this kind of bias affects algorithmic fairness and found that standard methods don’t work well to fix it. Instead, we came up with new ways to handle this type of missing data. Our results show that this type of bias can make things worse for some people, but our solutions help to make things more fair. |
