Measuring Fairness and Equity in Crowd-Flow Generation Models
The Team
Project Lead/s:
Afra Mashhadi, Assistant Professor of Computer Software and Systems at the University of Washington, Bothell.
Ekin Ugurel, PhD student at the University of Washington in Seattle.
Data Science Lead/s: Bernease Herman
DSSG Fellows: Apoorva Sheera, Jiaqi(Kiki) He, Manurag Khullar, Sakshi Chavan
Abstract or Executive Summary
Generative crowd-flow models, which simulate city population movements, have advanced from physics-based to neural network-based models, improving performance by incorporating city-specific features. However, concerns about the equity of these models and the potential social biases brought by the models remain largely unaddressed, which is critical for government planning, pandemic prevention, and etc. This project aims to develop and implement new fairness metrics for CF models to ensure equitable representation of each groups’ travel demands. The project involves exploring current equity disparities among different demographic groups, reviewing fairness literature, engaging with stakeholders, and creating a Python package to test CF models.
Expected final deliverables:
- A whitepaper describing group fairness metrics for spatial-temporal data;
- An open-source Python package implementing these metrics that can be used to test generative crowd-flow datasets before use by stakeholders.