Towards predicting student learning outcomes from learning management system interactions using machine learning

Thumbnail Image
Marcynuk, Kathryn L.
Journal Title
Journal ISSN
Volume Title
Advancements in classroom technology have resulted in new types of data collection in educational settings. Along with improvements in the fields of artificial intelligence and machine learning, this educational data can be used to study how we learn and create more personalised learning environments. Starting in March 2020 all in-person courses were abruptly moved to remote instruction in order to combat the COVID-19 pandemic. This influx of students taking remote courses presented a new opportunity to study how students interact with course materials. Remote learning courses at the University of Manitoba are offered using a learning management system (LMS) that centralizes all course activities and files and records user-activities. The use of machine learning techniques with education-based data is an emerging discipline that offers an opportunity to provide new insights in this area. This thesis presents a code-based tool to create student timelines from raw LMS date-time stamp data and extract features describing student behaviours within a single-term online course. The successes and limitations of these features to predict student grade outcomes were investigated using supervised and unsupervised machine learning models. The LMS data was also explored using neural network-based CNNs and transformers. The experiments presented in this thesis indicate that students predominately interact with the system at the same time on any given day relative to their previous interaction. The results further demonstrate that temporal features created from LMS interactions can predict student outcomes with greater than random accuracy. The neural network-based classifiers produced more accurate student outcome predictions than the feature-based ML models at the expense of interpretability. This thesis contributes to the body of knowledge on student modelling and prediction, as well as student behaviour within an LMS in an online course, and suggests that educators can help to reduce students' cognitive load and improve students' learning by updating the LMS at a consistent time of day.
machine learning, learning management system, student modelling, artificial intelligence, education, remote learning