Question and observed-coverage features retrospectively predict high-divergence questions with AUC 0.774, but the model is not validated for new-query routing¶

Claim · All claims
Evidence · 1 study · associational r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person · recheck by 2029: tool and model accuracy changes quickly

Subclaims¶

q2 i? Gradient boosting on question and observed-coverage features achieves test AUC 0.774 and average precision 0.570 for predicting top-quartile divergence among 441 questions. → Yubo Li 2026

Evidence¶

Yubo Li 2026¶

Yubo Li, Rema Padman, Ramayya Krishnan. (2026). Auditing Institutional Heterogeneity for Generative AI in Patient Education: A Large-Scale Study of 102 US Transplant Handbooks. arXiv preprint. https://arxiv.org/abs/2607.22606 doi:10.48550/arxiv.2607.22606

q2 · i? · associational · r2

Retrospective predictive analysis on one stratified 75/25 question-level split (seed 42) of the audited corpus, restricted to 441 questions with at least 30 non-absent pairs, of which 113 fall in the top divergence quartile. The article states there is no external-center or prospective validation.

"gradient boosting achieves test AUC 0.774 and average precision 0.570; logistic regression achieves 0.763 and 0.541."

Discussion¶

Related Claims¶

• Gradient boosting and extreme gradient boosting are the highest-performing classifiers for predicting enrollment, with balanced accuracy 0.91, F-score 0.88 and AUC 0.96 — related
• CatBoost achieved the best classification performance among tested algorithms for predicting course completion risk (F-measure .77, accuracy 78%, AUC .87) — related
• A random forest classified flagged versus non-flagged word problems with the best accuracy of five tested models (AUC = 0.75) — related
• Simple classifiers (Logistic Regression, Naive Bayes) outperformed tree-based models on this small dataset — related
• Baseline LR and XGBoost models achieve comparable predictive performance on OULAD, with XGBoost slightly higher AUC-ROC — related
• The deterministic pattern detector achieves high pair precision (73.21%) with explicit abstention but covers only 32% of real queries — related
• DeBERTa achieved the highest overall accuracy (86.36%) among the evaluated NLP models for classifying learner-generated questions, yet struggled with contextual nuances in complex queries — related
• Current-submission evidence and prior traces predict near-term persistent failure with moderate discrimination under student-disjoint evaluation — related