The Features of Students Paying and Not Paying Attention in Online Classes

The advent of online sessions in 2020, triggered by the global pandemic, gained widespread popularity due to its inherent flexibility. However, this convenience brought about challenges, notably in gauging student engagement. Prior research in gauging student engagement has focused on using computer vision to assess student attention. This has been limited, often encompassing only a few features, and lacking a comprehensive analysis of attentive or inattentive behavior during online sessions. To bridge this gap, this paper reports a study conducted to identify features of students paying and not paying attention in online classes. The study was based on gathering video recordings of online sessions and administering a questionnaire to discern instances of attention. The study identified ten attention behaviors and six non-attentive behaviors. Each behavior was then mapped to one or more computer vision techniques that could be used to extract features that identify the behavior in video sequences. The paper describes the extraction of features for two behaviors, leaning head on hand and head movement, not identified in previous work. The paper also reports on the application of three machine learning techniques, logistic regression, decision trees, and random forest classification, to determine whether the student is paying attention or not in online sessions based on a set of 8 features. The accuracy rates for these classifications were 79%, 83%, and 84%, respectively.