EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Composable Risk Certificates for Educational Prediction: Antichain Optimization and Controlled Simulations

András Szeberényi, Róbert Balku

Machine Learning and Knowledge Extraction · 07.11.2026 · 8 (11) · 370

DOI: 10.3390/make8110370 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Machine Learning and Knowledge Extraction, online erschienen am 07.11.2026. Verfasst von András Szeberényi und 1 weiteren Autor:innen. Die Arbeit nennt 84 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Risk-Budgeted Antichain Extraction (RBAE) selects disjoint regions from an independently trained tree by composing simultaneous regional error and probability-mass bounds. Its certificate controls accepted error across prespecified audit groups, without guaranteeing each rule separately. The theory establishes matched-local feasible-set inclusion, fixed-library value consistency under strict feasibility, and validity under restricted group-prior shift. Controlled simulations comprise six primary scenarios, a participation scenario, and a larger-library experiment with up to eight binary features. At a calibration size of 16,000, RBAE exceeds Local-CP coverage by 30.35 and 10.56 percentage points in two heterogeneous scenarios, but both Policy-CP and Threshold-CP achieve higher coverage in those same settings. At that size, near-boundary noise produces 999 empty policies in 1000 runs. Increasing label-flip probabilities by 0.05 causes group-limit violations in all 200 frozen policies in each of the two scenarios. Groupwise error control also coexists with a 19.27-point participation gap. The contribution is a compositional certificate for constrained rule selection, whose practical benefit depends on the permitted policy family and available calibration evidence. These experiments establish neither general superiority over direct policy calibration nor institutional deployment feasibility; educational effectiveness and human comprehension remain unevaluated.

Zitieren

APA

Szeberényi, A., & Balku, R. (2026). Composable Risk Certificates for Educational Prediction: Antichain Optimization and Controlled Simulations. Machine Learning and Knowledge Extraction, 8(11), 370. https://doi.org/10.3390/make8110370

Vancouver

Szeberényi A, Balku R. Composable Risk Certificates for Educational Prediction: Antichain Optimization and Controlled Simulations. Machine Learning and Knowledge Extraction. 2026;8(11):370. doi:10.3390/make8110370

BibTeX
@article{Szebernyi2026,
  author = {Szeberényi, András and Balku, Róbert},
  title = {Composable Risk Certificates for Educational Prediction: Antichain Optimization and Controlled Simulations},
  journal = {Machine Learning and Knowledge Extraction},
  year = {2026},
  volume = {8},
  number = {11},
  pages = {370},
  doi = {10.3390/make8110370}
}
Kostenlos starten →