A Temporal Information Sufficiency Framework for Prediction-Based Early Specialization Recommendation: An Uncertainty-Aware Cross-Cohort Evaluation
Year: 2026
Type: article
Abstract: Early prediction from progressively accumulated data raises a decision-timing question: when has enough information been observed to support a recommendation? We propose a temporal information sufficiency framework for prediction-based specialization recommendation and test whether the resulting decision time transfers across cohorts. Information saturation is declared when a simultaneous paired-bootstrap upper bound on the gain still obtainable from all later stages falls below a tolerance; predictive adequacy requires the lower performance bound to exceed a no-information baseline by a prespecified margin. Using 4627 transport construction engineering students from six cohorts at a Vietnamese university, 18 harmonized specializations, five nested curriculum-reference stages (19 to 94 credits), and expanding chronological evaluation, both conditions were met retrospectively only for two cohorts, at 58 and 39 credits; three cohorts reached saturation without adequacy. A decision frozen on one cohort and applied to the next never yielded an adequate stage. Diagnostics traced this instability to a restructured specialization portfolio, which led historical models to assign up to 64% of a cohort to discontinued specializations, and to curriculum revisions that made recorded course sets specialization-dependent. Saturation therefore does not imply adequacy, and a timing rule must be validated on future cohorts before it is used in advising.
Source: Computers
Authors: Van Long Nguyen, Thi Vinh Thanh Le
Language: English
Subfield: Computer Science Applications
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,172
OpenAlex ID: W7225738511