Early Predictions of Course Outcomes in a Flipped Classroom Context.

Among the different types of applications within the realm of learning analytics, predictors and early warning systems are particularly interesting thanks to their ability to identify potentially struggling students at an early stage of a course. This paper presents a performance comparison of different classification algorithms that aim to predict whether students will pass or fail a course, using only LMS activity data from its first few weeks. Two different time frames are considered: first, using exclusively data generated before any assessment activities are performed in the course; and second, allowing the use of data originating within the time period up to, but not including, the second partial exam of the course. The study targets a first-year engineering course at the University of Vigo. In the first scenario, the tested classifiers struggled to correctly identify passing students, an issue that is mitigated in the second scenario. The main hindrance to the classifiers' performance, particularly in the first analyzed scenario, is working with a set of unbalanced data in which the number of students who pass the course is extremely low compared to those who fail.