MACHINE LEARNING AND TEST ANXIETY: A CASE FOR RESTRAINED, TRANSPARENT AND STUDENT-GOVERNED USE
Test anxiety is a well-documented barrier to student performance, and machine learning (ML) is increasingly proposed as a way to predict, detect and respond to it. This position paper argues that, in the current state of evidence and regulation, ML should be used for test anxiety with restraint. Five problems are identified: weak and noisy ground-truth labels; evaluation practices, such as student-level data leakage and the absence of simple baselines, that overstate accuracy; unequal performance and base rates across groups; reliance on emotion inference from biometric signals that is scientifically contested and, in education institutions, prohibited in the European Union’s Artificial Intelligence Act with narrow exceptions; and the power asymmetries of assessment settings, where students cannot readily refuse and labels persist. The paper proposes a six-principle framework (support not sanction, student governance, simple models first, prospective and subgroup-aware validation, human in the loop, and no biometric emotion inference during assessment), and applies it to common proposals in a decision table. It addresses four objections and sets out a research agenda that includes trials of whether acting on predictions improves student outcomes, fairness analysis, and studies in low-resource contexts such as Nigerian schools.
Authors
- MNGUSUUR JAM
Publication Details
- Journal
- International Journal of African Innovation and Multidisciplinary Research
- Published
- 2026-11-22
- DOI
- https://doi.org/10.70382/mejaimr.v14i2.0155
- Primary Topic
- Online Learning and Analytics
- Type
- article
- Field-Weighted Citation Impact
- 0.00