Using multimodal learning analytics to model students' learning behavior in animated programming classroom

Studies examining students' learning behavior predominantly employed rich video data as their main source of information due to the limited knowledge of computer vision and deep learning algorithms. However, one of the challenges faced during such observation is the strenuous task of coding large amounts of video data through repeated viewings. In this research, we confirm the possibilities of classifying students' learning behavior using data obtained from multimodal distribution. We employed computer algorithms to classify students' learning behavior in animated programming classrooms and used information from this classification to predict learning outcomes. Specifically, our study indicates the presence of three clusters of students in the domain of stay active, stay passive, and to-passive. We also found a relationship between these profiles and learning outcomes. We discussed our findings in accordance with the engagement and instructional quality models and believed that our statistical approach will support the ongoing refinement of the models in the context of behavioral profiling and classroom interaction. We recommend that further studies should identify different epistemological frames in diverse classroom settings to provide sufficient explanations of students' learning processes.