Predicting Student Performance in a Non-linear Self-Paced Moodle Course (#1412)

Authors

Mercado, Jhon Fredy

Mendoza-Cardenas, Carlos

Gaviria-Gomez, Natalia

Fletscher, Luis

Botero, Juan F.

Abstract

Although student performance prediction in online courses has been extensively studied before, previous research efforts have focused mostly on courses with a linear structure, where the student is expected to take the lessons and assessments in a sequential (linear) and unique order. There are, however, non-linear courses where the student can take the lessons and assessments in any order they wish, which makes the performance prediction more challenging, as the cumulative assessment percentage might vary widely between students at a given point in time. Here, we present a data-driven method to predict student performance in non-linear courses. In particular, our method predicts whether or not a student will reach a final grade above a fixed threshold for different percentages of assessment completion. We use data from a Moodle course designed to prepare high-school students for the entrance exam of a public university. We show that for cumulative assessment percentages ≥ 60, our method has a mean F1-score above 70%. Finally, we also assess the importance of each feature used in the prediction, illustrating the effect of our data preprocessing and feature selection approach on the model performance.