Original title: Fairness aware learning analytics for identifying internet induced inequality in online examinations
Samer Yaghi, Aiman A. AbuSamra · September 2026
When students take online exams, test software assumes their internet works. But in places with weak connections, that assumption fails. A student whose web connection drops looks the exact same in the system as a student who did not study. Researchers wanted to fix this unfairness. They built a method using normal exam logs, which are digital records of student activity, to tell tech failures apart from poor preparation. The system uses a disruption score to spot genuine connectivity drops. They found that students who suffered bad disruptions scored just 10.0% during the drop, but earned 99.0% on a later, stable makeup test. Their past scores had averaged 77.8%, proving they knew the material. Practical hands-on exams suffered far more disconnections than theory tests. In some cases, students stayed logged in for entire sessions without any answers going through.
Imagine you're baking cookies from a recipe, but your oven keeps cutting out randomly. One batch, the oven dies for ten minutes, and your cookies come out doughy; another batch, you just forgot to set the timer and overbaked them, so if you only look at the final cookies both look equally bad. In this paper, the exams are the cookies, the internet connection is the oven, and the method uses connection records — not the cookie texture — to tell equipment failure from a baker's mistake.
Students in places with unreliable internet are being failed for equipment they don't control. This method reads ordinary exam logs to tell internet drops apart from genuine gaps in knowledge, so schools can correct those scores instead of punishing students for a lost connection.
The abstract doesn't say how the method handles intentional disconnections, whether it works with other exam platforms, or what the error rate is. It also doesn't specify a grade-adjustment policy; it only says the approach is a complement to existing institutional processes. The high confidence is based on just five students' before-during-after data, so larger validation is needed.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Fairness aware learning analytics for identifying internet induced inequality in online examinations — Samer Yaghi, Aiman A. AbuSamra — Discover Education — doi:10.1007/s44217-026-02182-6
Abstract licence: CC BY
Explanation generated by Synthesis from the abstract — it is not the original paper.