Machine Learning Driven Process Automation for Early Intervention on Course Noncompletion in Higher Education

Students who do not complete a course are often noticed only when final grades are published, too late to help. This paper proposes a framework that combines machine learning (ML) with process automation to find these students early and to decide, every week, which support action each student receives. The framework uses activity in the learning management system (LMS) to estimate each student’s risk of not completing the course and sorts students into action groups that match advisor capacity: the highest-risk students receive an automatic message and an advisor follow-up, medium-risk students an automatic message, and the others are monitored. A monitoring step signals when retraining is needed. The framework is tested on a recently published open dataset from KU Leuven covering two first-year courses over three academic years, counting students who never attempt the exam as not completing. Logistic regression, random forest, and gradient boosting are compared. The simple logistic regression model is as accurate as the complex models, more accurate in the first weeks, and flags at-risk students reliably within the first month. Advisor capacity strongly limits how many at-risk students are reached, and a model trained on earlier years remains useful for later ones.