810. Investigating stability of student-at-risk prediction model across years by means of XAI
Invited abstract in session TE-12: Learning Analytics, cluster Analytics and Data Science.
Tuesday, 16:15-17:45Room: FENP201
Authors (first author is the speaker)
1. | Elena Tiukhova
|
Faculty of Economics and Business, Research Center for Information Systems Engineering (LIRIS), KU Leuven | |
2. | Pavani Vemuri
|
LIRIS, KU Leuven | |
3. | Nidia Guadalupe Lopez Flores
|
Computer Science, Reykjavík University | |
4. | Anna Sigríður Islind
|
Reykjavik University | |
5. | María Óskarsdóttir
|
Department of Computer Science, Reykjavik University | |
6. | Stephan Poelmans
|
KU Leuven | |
7. | Bart Baesens
|
Decision Sciences and Information Mangement, K.U.Leuven | |
8. | Monique Snoeck
|
KU Leuven |
Abstract
Learning analytics (LA) is a fast-growing research domain focusing on modeling future learner outcomes, with student-at-risk prediction being among its most popular applications. According to Self-Regulated Learning (SRL) theory, a motivated learning choice can be represented with trace data collected by a Learning Management System (LMS), where higher-level indicators representing different aspects of learning can be engineered. Contemporary black-box predictive models using these indicators have limited usability towards explanatory LA. There is a difference between predictive and explanatory LA, with the former focusing on maximizing empirical precision at the cost of being in line with theory and having high transparency - more the focus of the latter. Our study aims to bridge the gap between predictive and explanatory LA modeling by combining advanced feature engineering based on SRL and predictive machine learning for student-at-risk prediction enriched with explainable AI (XAI) in an innovative way. Moreover, in the ever-changing contexts of learning, it is vital to ensure not only explainability, but also stability of models across time. Hence, we assess the robustness of models developed for a particular course across years using XAI techniques, extending existing research on stability of LA models and XAI in LA. Overall, our study proposes a novel approach to predictive LA incorporating explainability and robustness for long-term applicability.
Keywords
- Education
- Machine Learning
- Knowledge Engineering and Management
Status: accepted
Back to the list of papers