With the rise of courses in learning platforms, especially since the COVID-19 pandemic, universities face a major challenge: high dropout and failure rates. These courses require autonomy and more responsibility from learners, which can lead to difficulties for some. To face this challenge, it is essential to adopt appropriate strategies to identify and support at-risk students from the beginning of the semester. Although several studies have been conducted to identify these students, most of them focus on data collected at the end of the semester, making real-time intervention impossible. Additionally, the studies done during the semester are faced with a lack of relevant data or an insufficient volume of data for reliable predictions. Finally, a few studies explore the prediction of at-risk students with minimal data at the start of the semester. This work aims to identify key indicators (Background, engagement, and pre-test results) to detect at-risk students by modeling early learner profiles of 157 students and then predicting those at risk with minimal data from the beginning of the semester. Our results indicate that high-achieving students generally maintain their level of success, even if their engagement temporarily decreases. Engagement varies among students and evolves differently depending on learner’s profiles. The findings showed that most at-risk students belonged to lower-performing profiles who participated less in the first pre-test. In addition, students who participated in all pre-tests showed better final performance, except for some specific profiles. Finally, our analysis revealed that online engagement is not a strong indicator of high academic success and should be combined with other indicators.

Keywords: Online Learning Platforms (LMS), At-Risk Students, Early Prediction, Minimal Data, Pre-test.