| Summary: | While the recent technological advancements have enabled instructors to deliver mathematical concepts and theories beyond the physical boundaries innovatively and interactively, poor performance and low success rate in mathematic courses have always been a major concern of educators. More specifically, in an online learning environment, where students are not physically present in the classroom and access course materials over the network, it is toilsome for course coordinators to track and monitor every student’s academic learning and experiences. Thus, automated student performance monitoring is indispensable since it is easy for online students, especially those underperforming, to be “out of sight,” hence getting derailed and off-track. Since student learning and performance are evolving over time, it is reasonable to consider student performance monitoring as a time-series problem and implement a time-series predictive model to forecast students’ educational progress and achievement. This research paper presents a case study from a higher education institute where interaction data and course achievement of a previously offered online course are used to develop a time-series predictive model using a Long Short-Term Memory network, a special kind of Recurrent Neural Network architecture. The proposed model makes predictions of student status at any given time of the semester by examining the trend or pattern learned in the previous events. The model reported an average classification accuracy of 86 and 84% with the training dataset and testing dataset, respectively. The proposed model is trialed on selected online math courses with exciting yet dissimilar trends recorded.
|
|---|
