Skip to content

Toggle service links
  1. eSTEeM
  2. Projects
  3. Themes
  4. Supporting Students
  5. Disproved predictions of at-risk students: Some students fail despite doing well, others succeed despite predicted as at-risk

Disproved predictions of at-risk students: Some students fail despite doing well, others succeed despite predicted as at-risk

Project leader(s): 
Martin Hlosta
Faculty: 
STEM
Status: 
Current

The project will investigate the reasons why predictions provided by even sophisticated machine learning models sometimes do not correspond to reality. More specifically, we will examine the predictions of students that are at risk of not submitting or failing in their Tutor Marked Assignments (TMAs). These predictions are deployed via Early Alert Indicators (EAI) Dashboard and are available to Associate Lecturers (ALs) in undergraduate courses at the OU (more than 300 per academic year).

The goal of OUAnalyse predictions is to have actionable predictions, which can help ALs and other users with planning a potential intervention and ultimately to retain students. From our experience, there are more cases where data influencing students’ performance occur irregularly and it is not easy to be captured. For example, a student gets sick or a tutor intervenes at-risk student using a private phone.

Our goal is to provide more rigorous explanations of these cases. The results will be made available to ALs who can support the applications of predictive analytics to the teaching practice. For the cases, in which students were predicted as at-risk and there was a change in behaviour that leads to success, strategies used to retain at-risk students can be identified. These can be communicated to students at the start of the course or during the intervention. If features related to this change can be captured from the existing data, these will be used to feed to the learning algorithms. Implications of the proposed guidance on ALs practice can be an input for a subsequent project. This is an area that has been neglected in the existing learning analytics research, therefore this research will fill this gap.

Projects