MOOC Performance Prediction via Clickstream Data and Social Learning Networks

Posted date: 

In online learning, the ability to predict (in advance) whether a learner will be correct or incorrect in answering a question would be quite helpful to course instructors. For one, such methods would allow instructors to identify trouble areas for specific learners before they arise. In this work, we study student performance prediction for Massive Open Online Courses (MOOCs), where such analytics are particularly useful, given that learners typically answer only a fraction of the questions available for the course. By the same token, these analytics are also more challenging to obtain, given the sparsity of the question-response data. To combat this, the algorithms we develop leverage behavioral (clickstream) data collected about learners as they interact with course content to predict their performance. Tested against a large set of empirical data, we find that our schemes outperform standard algorithms (i.e., those without behavioral data) for all datasets and metrics tested. Moreover, the improvement is particularly pronounced when considering the first few course weeks, demonstrating the “early detection” capability of such clickstream data.

Link to article:  http://www.princeton.edu/~cbrinton/MOOC_perfPred.pdf