A predictive approach based on efficient feature selection and learning algorithms' competition: Case of learners' dropout in MOOCs

MOOCs are becoming more and more involved in the pedagogical experimentation of universities whose infrastructure does not respond to the growing mass of learners. These universities aim to complete their initial training with distance learning courses. Unfortunately, the efforts made to succeed in this pedagogical model are facing a dropout rate of enrolled learners reaching 90% in some cases. This makes the coaching, the group formation of learners, and the instructor/learner interaction challenging. It is within this context that this research aims to propose a predictive model allowing to classify the MOOCs learners into three classes: the learners at risk of dropping out, those who are likely to fail and those who are on the road to success. An automatic determination of relevant attributes for analysis, classification, interpretation and prediction from MOOC learners data, will allow instructors to streamline interventions for each class. To meet this purpose, we present an approach based on feature selection methods and ensemble machine learning algorithms. The proposed model was tested on a dataset of over 5,500 learners in two Stanford University MOOCs courses. In order to attest its performance (98.6%), a comparison was carried out based on several performance measures.