interaction data

Online learning programs have gained significant popularity in recent years. However, despite their widespread adoption, completion and success rates for online courses are notably lower than those for traditional in-person education. If students' final academic performance could be predicted early by analyzing their behavior within the virtual learning environment, timely alerts could be issued, and targeted interventions could be recommended to prevent underperformance and course abandonment. Previous studies have predicted academic performance using various features, such as demographic data, academic history, in-term exam results, and assignment assessments. However, many online learning platforms do not provide access to such data, rendering these methods ineffective. This study focuses on the early prediction of students' academic performance by extracting novel behavioral features based on their interactions with the online learning platform. To develop robust predictive models, we utilize an integrated approach combining multiple feature selection methods to extract the most informative interaction patterns, followed by application of advanced machine learning algorithms including ensemble learning techniques and artificial neural networks (ANNs). The evaluation results demonstrate that our proposed approach can predict students' final academic performance with an accuracy of 90.62%, using only data collected during the first third of the online course.