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

2023/01/10 by Denisa Gándara, Hadis Anahideh, Gándara, Denisa +4 · 3 citations
Mathematics · #Advanced Causal Inference Techniques #Computers and Society (cs.CY) #FOS: Computer and information sciences

paper · pdf · doi:10.48550/arxiv.2301.03784

openalex publication_date 2023/01/10 · openalex created_date 2023/01/12 · openalex updated_date 2026/08/01

Abstract

Colleges and universities are increasingly turning to algorithms that predict college-student success to inform various decisions, including those related to admissions, budgeting, and student-success interventions. Because predictive algorithms rely on historical data, they capture societal injustices, including racism. In this study, we examine how the accuracy of college student success predictions differs between racialized groups, signaling algorithmic bias. We also evaluate the utility of leading bias-mitigating techniques in addressing this bias. Using nationally representative data from the Education Longitudinal Study of 2002 and various machine learning modeling approaches, we demonstrate how models incorporating commonly used features to predict college-student success are less accurate when predicting success for racially minoritized students. Common approaches to mitigating algorithmic bias are generally ineffective at eliminating disparities in prediction outcomes and accuracy between racialized groups.

Cited by

• Learning Fair Representations with Kolmogorov-Arnold Networks
• Anticipatory Dignity and Digital Meritocracy: Philosophical Reflections on the Algorithmic Transformation of South Korea’s Education System
• Fairness under Competition

Related

• Evaluation of Fairness Trade-offs in Predicting Student Success
• Algorithms for College Admissions Decision Support: Impacts of Policy Change and Inherent Variability
• Predicting Early Dropout: Calibration and Algorithmic Fairness Considerations
• Should College Dropout Prediction Models Include Protected Attributes?
• Trading off performance and human oversight in algorithmic policy: evidence from Danish college admissions