Predicting Risk Earlier: Machine Learning to Support Success and Combat Inequity

• In Event: Assessing Student Progress and Attainment

Abstract

To target an early advising intervention more effectively, our research unit developed a machine learning model that predicts winter term retention before students matriculate. However, without thoughtful planning, machine learning can replicate – or even exacerbate – systemic biases experienced by past students. To combat this, we piloted a process of transparent, responsible machine learning performed by experts within our university. We worked closely with a range of stakeholders, including advisors and students, and rigorously reviewed the model for potential inequities. The model predicts winter retention three to five times as successfully as alternatives and works especially well for potentially vulnerable students. We view this as a successful use of machine learning to advance both equity and student success.

Authors

• Nathan P. Greenstein, University of Oregon
  Presenting Author

• Grant Crider-Phillips, University of Oregon
  Presenting Author