EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Mitigating algorithmic bias in educational prediction systems using reweighting and adversarial learning

William Villegas-Ch, Rommel Gutierrez, Joselin Garcia-Ortiz, Jaime Govea

Frontiers in Education · 30.07.2026 · 11

DOI: 10.3389/feduc.2026.1841724 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Frontiers in Education, online erschienen am 30.07.2026. Verfasst von William Villegas-Ch und 3 weiteren Autor:innen. Die Arbeit nennt 28 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The increasing integration of machine learning models into educational decision-making processes has raised concerns regarding algorithmic bias, particularly when sensitive attributes such as gender or socioeconomic status influence learning outcomes and assessment predictions. Models trained on historical educational data may reproduce structural inequalities, leading to disparities in automated decisions across student groups. This study presents an empirical fairness-evaluation framework for educational prediction systems, comparing a baseline classifier with demographic reweighting and adversarial debiasing. The approach is evaluated through statistical testing, subgroup analysis, and robustness assessment using standard fairness metrics, including Statistical Parity Difference (SPD), Equal Opportunity Difference (EOD), and Disparate Impact (DI), together with predictive indicators such as accuracy and F1-score. Experimental results show that mitigation effects are dataset- and metric-dependent. In ASSISTments, the adversarial configuration improves predictive performance but increases SPD and EOD relative to the baseline, whereas reweighting produces nearly identical behavior to the unmitigated model. In OULAD, the adversarial configuration slightly reduces predictive performance while widening the observed fairness disparities. Under Gaussian perturbations ( σ = 0.05 ), the adversarial model preserves most of its predictive capacity. Overall, fairness-aware mitigation should be evaluated as a trade-off between predictive utility and subgroup equity rather than as a single optimization objective, contributing to more transparent educational decision-making in learning analytics systems.

Zitieren

APA

Villegas-Ch, W., Gutierrez, R., Garcia-Ortiz, J., & Govea, J. (2026). Mitigating algorithmic bias in educational prediction systems using reweighting and adversarial learning. Frontiers in Education, 11. https://doi.org/10.3389/feduc.2026.1841724

Vancouver

Villegas-Ch W, Gutierrez R, Garcia-Ortiz J, Govea J. Mitigating algorithmic bias in educational prediction systems using reweighting and adversarial learning. Frontiers in Education. 2026;11. doi:10.3389/feduc.2026.1841724

BibTeX
@article{VillegasCh2026,
  author = {Villegas-Ch, William and Gutierrez, Rommel and Garcia-Ortiz, Joselin and Govea, Jaime},
  title = {Mitigating algorithmic bias in educational prediction systems using reweighting and adversarial learning},
  journal = {Frontiers in Education},
  year = {2026},
  volume = {11},
  doi = {10.3389/feduc.2026.1841724}
}
Kostenlos starten →