Agentic data science with the Data Agent Kit
cloud.google.com·6h ago
AI Research
arXiv CS.AI·Sep 1·5 min read
VFR-Audit: Verdict-Level Reliability for Fairness Audits in Hospital Length-of-Stay Prediction
TL;DR
VFR-Audit introduces a framework for assessing the reliability of fairness audits in hospital length-of-stay predictions, focusing on verdict stability. It quantifies the likelihood of verdict changes using the Verdict Flip Rate (VFR) and evaluates reliability across multiple dimensions.
✦ Why It Matters
Implement VFR-Audit to enhance the reliability of fairness assessments in your hospital's AI systems today.
Key Takeaways
- Utilize VFR-Audit for reliable fairness assessments in AI models.
- Measure verdict stability using the Verdict Flip Rate (VFR).
- Evaluate bias-mitigation impacts on audit outcomes effectively.
- Standardize fairness audits across different hospital settings.
How It Works
VFR-Audit employs stratified bootstrap resampling to calculate the Verdict Flip Rate (VFR), which indicates the probability of a verdict changing. This method allows for a more nuanced understanding of verdict stability by assessing how different factors, such as cohort size and hospital variability, influence audit results.