Abstract
Massive Open Online Courses (MOOCs) have been suffering a very level of low course certification (less than 1% of the total number of enrolled students on a given online course opt to purchase its certificate), although MOOC platforms have been offering low-cost knowledge for both learners and content providers. While MOOCs discussion forums' rich numeric and textual data are typically utilised to address many MOOCs challenges, e.g., high dropout rate, identifying intervention-needed learners, analysing learners' forum discussion and interaction to predict certification remains limited. Thus, this paper investigates if MOOC discussion forum-based data can predict learners' purchasing decisions (certification). We use a relatively large dataset of 23 runs of 5 FutureLearn MOOCs for temporal (weekly-based) prediction, achieving promising accuracies in this challenging task: 76% on average, across the five courses.