Learning Analytics for Predicting Student Performance in Online Learning Environments

The fast-growing numbers of the online learning space have led to the storage of huge amounts of student-to-student interaction data in Learning Management Systems (LMS). However, very often, the educational institutions do not have systematic systems of the usage of such data to help to identify students who are at risk of the future changes in time. This paper fills this gap by building and testing predictive models to predict academic performance of students through learning analytics. Using a quantitative research design, we studied the interaction logs, assessment data, and recorded engagement of 350 university students taking a course all semester-long through Moodle. The most essential behavioral variables, such as the number of logins, the timeliness of submission of assignments, success of discussion forums and watching video lectures were extracted and were used to train and compare various machine learning models, namely, Logistic Regression, Random Forest, and Support Vector Machines. Accuracy, precision, recalls and F1-score were used to measure model performance. Findings indicate that the highest predictive accuracy is experienced in the Random Forest (87-percent), and the assignment submission pattern and a regular frequency of logging into the account are the most potent predictors of ultimate academic achievement. These findings highlight the possibility of learning analytics to support early warning systems based on data, which is why early pedagogical interventions can be provided. This paper becomes a contribution to the literature on educational data mining through the empirical evidence of the relationships between behavioral indicators based on conventional LMS logs and their good predictive abilities of student results, which would provide practical implications to teachers, instructional designers, and institutional policymakers seeking to increase student learning and to tailor support in online learning settings.