Real-Time Prediction of Students' Activity Progress and Completion Rates

In classrooms, some transitions between activities impose (quasi-)synchronicity, meaning there is a need for learners to move between activities at the same time. To make real-time decisions about when to move to the next activity, teachers need to be able to balance the progress of their students as they work at different paces. In this paper, we present a set of estimators that can be used in real time to predict the progress and completion rates of students working on computer-supported activities that can be divided into sequential subtasks. With our estimators, we investigate what effect the average progress rate of the class, a given number of previous steps, or weighting the proportion of progress assigned to each subtask has on predictions of students' progress. We find that accounting for the average class progress rate near the beginning of the activity can improve predictions over baseline. Additionally, weighted subtasks decrease prediction accuracy for activities where the behaviour of faster students diverges from the average behaviour of the class. This paper contributes to our ability to provide accurate student progress predictions and to understand the behaviour of students as they progress through the activity. These real-time predictions can enable teachers to optimize learning time in their classrooms.