A new hybrid approach to detect and track learner's engagement in e-learning

Learner engagement is a critical concept that can lead to satisfaction, motivation, and success in e-learning courses. It covers contextual, emotional, behavioral, cognitive, and social aspects. The instructors have difficulties identifying who is involved in the courses and the lack of face-to-face interaction with a learning resource to act upon and reduce the dropout rate. This paper presents a novel approach that aims to predict learner engagement in online courses and quantify the relationship between the learners' success and their engagement. For this purpose, we used the traces gathered from 1 356 learners' reactions in e-learning courses during the winters of 2020, 2021, and 2022, to implement this approach. To model the learning engagement, a variety of features were considered such as the total number of posts made in forums and the total time spent on the e-learning platform. This study uses the BiLSTM method with FastText word embedding to detect learners' emotions from the forum discussions. Then, an unsupervised clustering technique based on the new dataset was used to cluster learners into groups according to their engagement level. Several supervised classification algorithms were trained and their performance was evaluated using cross-validation techniques and diverse precision metrics. The findings indicate that the decision tree rule model is more relevant than others, with an accuracy of 98% and an AUC score of 0.97. The conclusions of this research reveal that most learners are observers and that there is a nonlinear correlation between learning success and learning engagement.