Performance prediction in online academic course: a deep learning approach with time series imaging

With the COVID-19 outbreak, schools and universities have massively adopted online learning to ensure the continuation of the learning process. However, in such setting, instructors lack efficient mechanisms to evaluate the learning gains and get insights about difficulties learners encounter. In this research work, we tackle the problem of predicting learner performance in online learning using a deep learning-based approach. Our proposed solution allows stakeholders involved in the online learning to anticipate the learner outcome ahead of the final assessment hence offering the opportunity for proactive measures to assist the learners. We propose a two-pathway deep learning model to classify learner performance using their interaction during the online sessions in the form of clickstreams. We also propose to transform these time series of clicks into images using the Gramian Angular Field. The learning model makes use of the available extra demographic and assessment information. We evaluate our approach on the Open University Learning Analytics Dataset. Comprehensive comparative study is conducted with evaluation against state-of-art approaches under different experimental settings. We also demonstrate the importance of including extra demographic and assessment data in the prediction process.