Classification of live course watching status of university students with the transfer learning method

Distance education eliminates people's disadvantages, such as time and space. It is a strategy that gives individuals major advantages in maintaining the consistency of their learning processes. With this process, via both synchronous and asynchronous channels, learners can be reached. Among the synchronous processes, the tasks used involve live lessons. The lecturers and learners come together at the same time on the internet with live classes, and the lessons are taught. Learning processes are important, particularly in live lessons. The aim of this research is to investigate the actions and conduct of students attending live lessons. Thus, it is possible to assess the participation rate of students in live lessons. The research was performed using quantitative techniques. In the analysis, the obtained data is presented as percentage and frequency. In this case, two live lectures for a specified course are scheduled to be registered. The participants who wanted to voluntarily participate in the study were told that during the live lesson they had to leave their cameras open. The two collected live lecture recordings were divided into intervals of five seconds. With 17,036 images, which were revealed from these images, labeling work has been completed. The conduct of students following the lesson differs throughout the lesson, according to the results obtained from the report. In this sense, the proportion of learners watching the course during the lessons has changed. The rate of watching the lesson was decided to be the highest, especially at the beginning and the end of the lesson. It was noted that during the classes, three out of every four participants followed the course on average. It was concluded that by answering questions or laughing, some of the respondents attended both classes and engaged in the lecture. Participants' actions not to watch the lesson were to look the other way, bow their heads, and leave the field of the camera. To classify the situation of listening or not listening to the lesson, deep learning-based transfer learning models were used and a feature vector with 1000 columns was created. With the Cubic SVM classification algorithm and MobileNetv2 model, 92.0% successful classification was achieved. Among the study recommendations, it was reported that it would be advantageous for class participation to increase the degree of engagement in the lessons.