Predicting Peer-Review Participation at Large Scale Using an Ensemble Learning Method

Abstract

Peer review has been an effective approach for the assessment of mas-sive numbers of student artefacts in MOOCs. However, low student participation is a barrier that can result in inefficiencies in the implementation of peer reviews, disrupting student learning. In this regard, knowing earlier the estimate number of peer works that students will review may bring numerous pedagogical utilities in MOOCs. Previously, we have attempted to predict student participation in peer review in a MOOC context. Building on our previous work, in this study we pro-pose an ensemble learning approach with a refined set of features. Results show that the prediction performance improves when a preceding classification model is trained to identify students with no peer-review participation and that the re-fined features were effective with more transferability to other contexts.