Utilizing Student Time Series Behaviour in Learning Management Systems for Early Prediction of Course Performance

Predictive analytics in higher education has become increasingly popular in recent years with the growing availability of educational big data. Particularly, a wealth of student activity data is available from learning management systems (LMSs) in most academic institutions. However, previous invest...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:Journal of Learning Analytics Ročník 7; číslo 2; s. 1 - 17
Hlavní autoři: Chen, Fu, Cui, Ying
Médium: Journal Article
Jazyk:angličtina
Vydáno: Society for Learning Analytics Research 01.01.2020
Témata:
Foreign Countries
Grade Prediction
Integrated Learning Systems
Learning Analytics
Student Behavior
Time on Task
Undergraduate Students
Canada
ISSN:1929-7750, 1929-7750
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:Predictive analytics in higher education has become increasingly popular in recent years with the growing availability of educational big data. Particularly, a wealth of student activity data is available from learning management systems (LMSs) in most academic institutions. However, previous investigations into predictive analytics in higher education using LMS activity data did not adequately accommodate student behaviours in the form of time series. In this study, we have applied a deep learning approach--long short-term memory (LSTM) networks--to analyze student online temporal behaviours using their LMS data for the early prediction of course performance. To reveal the potential of the deep learning approach in predictive analytics, we compared LSTM networks with eight conventional machine-learning classifiers in terms of the prediction performance as measured by the area under the ROC (receiver operating characteristic) curve (AUC) scores. Results indicate that using the deep learning approach, time series information about click frequencies successfully provided early detection of at-risk students with moderate prediction accuracy. In addition, the deep learning approach showed higher prediction performance and stronger generalizability than the machine learning classifiers.
ISSN:1929-7750
1929-7750
DOI:10.18608/JLA.2020.72.1