Cross-course generalizability of SRL-aligned predictive models using digital learning traces

Computers & Education 2026 Open access

Read full text
Authors: Jakob Schwerter, Loreen Sabel, Judith Bose, Matthew L. Bernacki, Di Xu, Marko Schmellenkamp, Thomas Zeume, Philipp Doebler
DOI: 10.1016/j.compedu.2026.105670
Cited by: 1
References: 58

Abstract

• SRL-aligned digital trace indicators enable early prediction of exam risk in theory-intensive computer science courses. • Predictive models generalize only partially across courses and institutions, with substantial performance loss under base-rate shifts. • Elastic Net models generalize more robustly than high-performing ensemble methods despite lower in-sample accuracy. • Probability calibration improves the interpretability of predicted risk but cannot fully compensate for cross-institution base-rate differences. • Stable SRL-related behavior patterns predict success even when their week-level manifestations vary.

Fields of Study & Subjects

Model Reduction and Neural Networks Student Assessment and Feedback Model-Driven Software Engineering Techniques

Publication details

Journal / source
Computers & Education
Publisher
Elsevier BV
Volume
253
Pages
105670–105670
Publication date
26 May 2026
Language
EN