Student learning analysis in e-learning using deep neural networks

In a period where demand for web-based education systems is rapidly increasing, estimating whether a student is learning or not and how a course can be adapted to the needs of its students becoming crucial. Flow theory provides an explanation for how the emotional state of an individual can have an impact on their learning. Hence, in the scope of e-learning, estimating the flow states of students can provide useful information that will lead to the estimation of their learning. The challenge in flow state estimation in e-learning platforms is to extract parameters that reflect the effort, the activity, and the performance of students. In this scope, this thesis proposes a deep neural network and flow theory-based method aiming to estimate student learning in a course taught with e-learning. First, the activities of the students are analyzed using the interaction of students with an e-learning platform that comprises classical e-learning pages and a timeline tool. This analysis contains activity heatmaps and deep neural network techniques. Then, the performances of the students are obtained through quizzes presented with the e-learning interface. The activities and performances are then validated with a statistical analysis using student surveys. Using the activities and the performances the method generates an estimation about the flow state of students in a course. The proposed deep neural network model acts as a dimension reduction method that extracts patterns from the activity heatmaps. The student profile biases are eliminated by conducting experiments in two different courses from two different disciplines. The proposed model presents superior performance over the dimension reduction methods used in the literature in terms of correlation with the surveys.