Using online problem-based learning networks to understand and predict performance

Across four online problem-based learning courses, interaction networks were linked with grades through correlations and regression, then checked on the next year's course data. Denser group interaction, cohesion, and ties to prominent peers were associated with performance, and a classifier reported 93.3 percent accuracy. The study shows how relational features can add context to early-support models, while its interpretation also warns that an alert should trigger assistance rather than label a learner as deficient.

Moderate-to-strong associations appeared across the studied courses, and the next-year validation supported the selected network predictors; tutor interaction was negatively associated with grades, plausibly because tutors responded more to struggling groups. The courses came from one educational setting and the models use observational platform traces rather than randomized exposure. Prediction accuracy depends on class balance, threshold, course design, and future similarity; correlation and prediction do not establish that changing centrality would change grades.