Predicting student performance over time. A case study for a blended-learning engineering course

Abstract

In recent years, different studies have focused in analyzing whether it is possible to explain and predict performance of students based on information we know about them, and in particular, on that obtained from Learning Management Systems (LMSs). A review of existing literature shows we can still raise no conclusion, and in particular when dealing with face to face (F2F) studies. In this article, we analyze the performance of a first-year engineering course, offered in a higher education institution (a public university). The course under analysis lasts for 12 weeks and is offered with flipped classroom methodology. Activities that students should follow out of class are scheduled in advance, and communicated to students during the learning period. In addition, there has been a previous effort to align learning activities and learning outcomes. The goal is to determine if prediction models fed with data gathered during the learning process can provide an accurate estimator of students at risk. This risk evaluation will be done considering as core data those reflecting activity, being of particular relevance, traces stored in LMS as part of the learning process. Our study demonstrates performance can be estimated based on this data, with increasing accuracy over time. Activity performed by the student is linked to academic result, and this relation is verified even when not taking into account any graded results obtained during the learning process.