ARTFEED — Contemporary Art Intelligence

LEAP Protocol Prevents Temporal Leakage in LMS Early-Warning Models

other · 2026-05-26

A new study from arXiv introduces LEAP (Leakage-Excluded Early-Availability Protocol), a method to prevent temporal leakage in early-warning models built from Learning Management System (LMS) logs. Temporal leakage inflates reported early performance by using future information not available at prediction time. LEAP enforces cutoff-first truncation before joins and aggregation, and audits feature provenance to exclude post-cutoff evidence. The protocol is instantiated on the Open University Learning Analytics Dataset (OULAD) across weekly cutoffs. Standard learning methods are evaluated using ROC-AUC, PR-AUC, Brier score, and F1@0.5. The work formalizes cutoff-based early outcome prediction under a temporal availability constraint.

Key facts

  • LEAP stands for Leakage-Excluded Early-Availability Protocol
  • Temporal leakage occurs when prediction uses information unavailable at prediction time
  • LEAP enforces cutoff-first truncation prior to joins and aggregation
  • The protocol audits feature provenance to prevent post-cutoff evidence
  • Evaluated on the Open University Learning Analytics Dataset (OULAD)
  • Evaluation metrics include ROC-AUC, PR-AUC, Brier score, and F1@0.5
  • The study formalizes cutoff-based early outcome prediction
  • Published on arXiv with identifier 2605.25794v1

Entities

Institutions

  • arXiv
  • Open University

Sources