Predicting Student Performance Based on Logs in Moodle LMS

Abstract

Context: This innovative practice full paper presents a methodology to predict at-risk students in the context of a course assisted by an LMS (Learning Management System). LMSs generate large amounts of data about courses and students, which allows schools to make useful insights with the help of computational analytical tools. Most educational institutions claim that the most significant issue in virtual learning is high student dropout rates, and school performance is one of its main factors. Objective: Our study aims to use Machine Learning techniques based on logs from the Modular Object-Oriented Dynamic Learning Environment (Moodle). Those data are used to analyze student behavior and create a model that helps detect students at risk. Method: This paper used institutional data and trace data generated by LMS of a Computing education technical courses, blended and distance learning, at high school. We compared 7 algorithms with models trained at 6%, 20%, 40%, and 60% of the course duration, with the intent of exploring the compromise between early and late detection of at-risk students. Our model has 69% positive classe (failed) and 31% negative class (passed), and the false positives cost is important. Results: The results show 7 created models of predicting. The findings for Random Forest performed the best when predicting a student's performance. Conclusion: Our study provides a student at-risk prediction model using ML techniques on logs in Moodle LMS and may guide future studies and tool development to reduce these high dropout rates.