Fairness aware learning analytics for identifying internet induced inequality in online examinations

Abstract

Abstract Online examinations delivered through learning management systems implicitly assume stable internet connectivity — an assumption that fails systematically in infrastructure-constrained settings. In such environments, a student who loses connectivity mid-examination can be recorded identically to one who is academically unprepared, producing a measurable form of assessment inequity. This study presents a fairness-aware learning-analytics approach for identifying internet-induced inequality directly from standard examination logs. We analyse 675 examination instances from 226 students across five sections and seven assessments at the University College of Applied Sciences, Gaza, Palestine, treated as a stress-test case for connectivity-constrained assessment. A severity-graded disruption definition separates substantial infrastructure-induced failure from incidental omissions. Seven students flagged as substantially disrupted later sat a stable supplementary examination; the five with an unimpeded prior assessment had averaged 77.8% before disruption, scored 10.0% under it, and 99.0% afterwards, returning to rather than exceeding their own baseline (rank-biserial r = 1.00; bootstrap 95% CI for the gain [78.0, 97.1] points). The result is invariant across every disruption threshold from r ≥ 0.30 to r ≥ 0.70. Prevalence differs sharply by format, reaching 34.5% in practical examinations against 1.5–4.4% in theory examinations, and a distinct complete-disconnection signature appears in which students remain in full-length sessions with no transmitted answers. The findings indicate that internet-induced inequality is detectable, quantifiable, and separable from academic performance, offering a reproducible complement to the institutional processes that already exist.