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

This study investigates bias in college student success prediction models, focusing on racial disparities. Using the Education Longitudinal Study of 2002, the researchers evaluate the fairness of several machine learning models (Random Forest, SVM, Logistic Regression, Decision Tree) and the effectiveness of bias mitigation techniques (reweighting, Disparate Impact Remover, Exponentiated Gradient Reduction, Meta Fair Classifier). They find significant bias against Black and Hispanic students across various fairness metrics (statistical parity, equal opportunity, predictive equality, equalized odds) and limited success in mitigating this bias using existing techniques.

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