A decision-aware framework for pre-regulatory academic early warning under resource constraints

Abstract

Academic early-warning systems are commonly evaluated as prediction tools, although their institutional value depends on how risk estimates are translated into support decisions. We developed and retrospectively evaluated a decision-aware framework for pre-regulatory academic warning using longitudinal administrative records spanning six academic years and twelve standardised semesters at a regional public university. Condition-specific risks of violating formal academic-warning regulations were estimated using three predictive backbones and evaluated through semester-based rolling-origin testing. Calibrated probabilities were converted into support priorities using context-sensitive expected regulatory violation costs under a fixed semester-level support budget. At 20% intervention coverage, the proposed policy reduced the missed intervention rate to 0.447, compared with 0.514 for the strongest non-ERVC comparator, while increasing positive predictive value from 0.377 to 0.429. The estimated equity gap in missed intervention was smaller under ERVC (0.052 versus 0.081 for the strongest non-ERVC comparator), but the paired difference did not reach conventional statistical significance ( \(p=0.06\) ), indicating a favourable but statistically uncertain equity pattern. These findings support evaluating early-warning systems through decision quality, resource use, and equity, while preserving human oversight and formal academic governance.

The authors' abstract, as published at the source. Scientific Reports, 2026 · DOI ↗