[arXiv]score: 0.53
LEAP Protocol Exposes Temporal Leakage Inflating Early Warning Model Performance on OULAD
May 26, 2026
A cutoff-first temporal availability protocol (LEAP) applied to the OULAD dataset shows that standard LMS early-warning pipelines suffer from temporal leakage that artificially inflates reported ROC-AUC, PR-AUC, and F1 scores.
cs.AI
HOW THIS AFFECTS YOU
●
builderYou can apply LEAP's cutoff-first truncation and feature provenance audit to validate that your early warning or churn prediction pipelines aren't reporting inflated metrics.
●
researcherLEAP formalizes a reproducible leakage-exclusion protocol for time-series educational prediction that likely applies broadly to any event-log-based early warning system.