Improving Prediction of Student Performance in a Blended Course

Abstract

Traditionally, systems supporting blended learning focus only on one portion of the course by tracing students’ interaction with learning content at home. In this paper, we argue that in-class activity can be also instrumental in eliciting the true state of students’ knowledge and can lead to more accurate models of their performance. Quizitor is an online platform that delivers both the at-home and the in-class assessment. We show that a combination of the two streams of data that Quizitor collects from students can help build more accurate models of students’ mastery that help predict their course performance better than models separately trained on either of these two types of activity.