A recurrent neural networks based framework for at‐risk learners' early prediction and MOOC tutor's decision support

Since their beginning, Massive Open Online Courses (MOOC) have known great success and have managed to establish themselves with significant enrollment rates. However, this success was quickly disrupted by the drop‐out phenomenon observed in the majority of MOOCs, which reaches 90% in some courses. Studying and understanding this phenomenon, and consequently determining the relevance of the efforts made to develop MOOCs, has led several researchers to propose predictive models of learners at risk of dropping out. On one hand, these models have been made relying on machine learning and the massive data generated by learners' navigation. On the other hand, these models only provide weekly predictions and do not give clear visibility about the overall course progress. We present in this paper a framework based on the recurrent neural networks' strengths which uses generator and predictor modules. Our framework allows not only the prediction of dropouts but also the generation of each learners' behaviors during the whole course since its first week. Besides, an OLAP analytical module proved great support for MOOC moderators to report on the learners' behavior at‐risk to target their interventions and guide their support.