14 June 2021
by Kobi Gal, University of Edinburgh
The remote collaborative activities culminated this month with the submission for the undergraduate thesis by Stefi Tirkovi on Massive Online Learning Courses (MOOCs).
The shift from traditional face-to face learning to online activities forced by the COVID-19 pandemic has shown an immense increase in enrolment rates in MOOC platforms. This underpins the importance of open online courses and how crucial research in that realm is in order to gain deeper understanding about users, their needs and behaviours to aid the future development of distance education.
Her research project makes the following contributions:
- Formulation of two different computational tasks concerning student performance. The first aimed to output a predicted grade, and the second one presents a binary prediction problem of whether an assignment submission will take place.
- Feature design based on multitude of data sources. Different features clustered in “feature families” were identified in order to perform the tasks above. Each of them includes data from one of the three different sources – clickstream logs, discussion forum and previous assessment.
- Design of predictive models using the identified feature sets for each of the formulated tasks. After extracting the defined features from the raw data, linear and logistic regression algorithms were utilised to predict performance for each of the two respective problems.
- Evaluation of the designed models and the predictive power of the feature sets. An extensive evaluation and comparison between all models is presented using data from two Edinburgh MOOCs. The models are compared with respect to predefined baselines and between each other. Moreover, the best feature combinations are further explored and optimised using other ML algorithms. Stefi’s results demonstrated the benefit of using predictive models for all of these tasks.