EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC 4.0

Resource-Aware Fair Conformal Prediction for Educational Early Warning Systems

Mira Chen, Avery Liu, Ryan Rahman

Journal of Educational Development and Social Sciences · 30.10.2026 · 1 (6) · 71-86

DOI: 10.62051/jedss.v1n6.08 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Educational Development and Social Sciences, online erschienen am 30.10.2026. Verfasst von Mira Chen und 2 weiteren Autor:innen. Die Arbeit nennt 32 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Early warning systems (EWS) in schools and universities operate under hard resource limits: counselors can review only a fixed number of students each term, and support programs have a fixed number of places. The equalized conformal classification (ECC) layer bounds, within every demographic group, how often at-risk students are missed and how often on-track students are flagged, but its counselor workload is an uncontrolled by-product of the chosen error level. This paper proposes RA-ECC, a resource-aware conformal triage rule that reverses this relation. RA-ECC computes Mondrian conformal p-values over group-by-outcome cells, refers exactly the K students whose less plausible outcome is still the most plausible, and resolves empty prediction sets automatically, so the review capacity is met by construction and all group-by-outcome error rates share one capacity-implied level. We prove that empty-set resolution preserves the ECC error bounds, that a common level is minimax-optimal among group-wise threshold triage policies, and that a Dvoretzky–Kiefer–Wolfowitz–Massart argument certifies the data-dependent level uniformly, which also supports joint planning of review and support capacity. On three public U.S. cohorts (Tennessee STAR, High School and Beyond, and the LSAC Bar Passage Study; 30 random splits each), our reproduction of ECC matches the published coverage to within 0.01, and at a 20% review capacity RA-ECC lowers the worst-group automated error relative to confidence-based deferral from 0.375 to 0.147, from 0.310 to 0.235, and from 0.510 to 0.171. RA-ECC never exceeds capacity, whereas a level fixed in advance does so in 30–40% of cohorts, and its mean cell errors stay within 0.008 of the capacity-implied level.

Zitieren

APA

Chen, M., Liu, A., & Rahman, R. (2026). Resource-Aware Fair Conformal Prediction for Educational Early Warning Systems. Journal of Educational Development and Social Sciences, 1(6), 71-86. https://doi.org/10.62051/jedss.v1n6.08

Vancouver

Chen M, Liu A, Rahman R. Resource-Aware Fair Conformal Prediction for Educational Early Warning Systems. Journal of Educational Development and Social Sciences. 2026;1(6):71-86. doi:10.62051/jedss.v1n6.08

BibTeX
@article{Chen2026,
  author = {Chen, Mira and Liu, Avery and Rahman, Ryan},
  title = {Resource-Aware Fair Conformal Prediction for Educational Early Warning Systems},
  journal = {Journal of Educational Development and Social Sciences},
  year = {2026},
  volume = {1},
  number = {6},
  pages = {71--86},
  doi = {10.62051/jedss.v1n6.08}
}
Kostenlos starten →