Following freshman footprints: early student activity & academic performance

Study performance in the first semester is highly predictive for study success. Hence, early interventions targeted at students-at-risk are called for. This study examines the potential of first-semester system activity data for student at-risk classification. By incorporating login data from the learning management system Moodle and information on registered exams, we aim to improve outcome forecasts even before exam results are available. Using machine learning techniques, we demonstrate that including this data significantly enhances classification accuracy for all three outcomes: dropout, (in-) activity, and first semester retention. Moreover, classifying student activity levels can be done under stricter data availability constraints, making it a potential alternative to traditional study success or failure predictions. Our regression analysis reveals a strong relationship between learning platform activity and study performance. Additionally, it suggests that learning activity later in the semester is more important for academic performance than activity during the first month.