This paper explores the prediction of student performance characteristics based on their interactions with a learning management system (LMS). By leveraging extensive log data from these interactions, we employ various Machine Learning (ML) techniques to train and compare multiple predictive models. Our primary objective is to forecast the final grades of students in a given course, though our methodology is versatile and can be adapted to predict other performance metrics. The study covers the entire process, including the preprocessing of raw interaction logs to filter and clean the data, the extraction of meaningful features that capture student behavior, and the application of several ML algorithms, including logistic regression, gradient boosted trees, and neural networks. We compare the accuracy of these models to identify the most effective approach for predicting student performance. The findings of our research reveal that specific patterns of interaction, such as regularity in accessing course materials and participation in discussion forums, are strong indicators of academic success. These insights offer valuable guidance for educators and administrators, enabling them to design targeted interventions and support mechanisms to enhance the online learning experience, ultimately fostering better educational outcomes for students.

Keywords: Learning Management System, Machine Learning, student performance prediction.