Inside the Black Box: Detecting and Mitigating Algorithmic Bias across Racialized Groups in College Student-Success Prediction

Colleges and universities increasingly use predictive algorithms to inform decisions related to admissions, budgeting, and student-success interventions. Because such algorithms rely on historical data, they can encode and reproduce societal injustices, including racism, leading to biased predictions for racially minoritized students. This study models bachelor’s degree attainment using several machine-learning approaches and evaluates leading bias-mitigation techniques. Using nationally representative data from the Education Longitudinal Study of 2002, we demonstrate that models built on commonly used features for predicting college-student success produce racially biased results and that commonly used mitigation techniques have limited effectiveness in reducing these biases.

Hadis Anahideh, Matthew P Ison, Anuja Tayal, Denisa Gándara