Fairness Audit Reveals Bias in College Early Warning System Against Older and Female Students
An audit focused on fairness within a machine learning pipeline at Centennial College uncovered significant misallocation issues in its Early Warning System (EWS). This evaluation, outlined in arXiv preprint 2604.19468v1, examined inequalities related to gender, age, and residency. By replicating the EWS model with institutional data, researchers discovered that younger, male, and international students were disproportionately identified for support, while older and female students with similar dropout risks were often overlooked. The post-processing phase exacerbated these inequalities by categorizing probability scores. This study is part of a long-term collaboration with Centennial College and builds on previous efforts that introduced the ASP-HEI Cycle framework, emphasizing the importance of fairness audits in educational machine learning systems.
Key facts
- Fairness audit conducted on deployed Early Warning System at Centennial College
- arXiv preprint 2604.19468v1 documents the audit findings
- Audit evaluated disparities by gender, age, and residency status
- Younger, male, and international students disproportionately flagged for support
- Older and female students with comparable dropout risk under-identified
- Post-processing procedures amplified existing disparities
- Research builds on multi-year collaboration and prior ethnographic work
- Study used replica-based audit approach with institutional training data
Entities
Institutions
- Centennial College