Abstract:

In the Learning Management System(LMS), each class is assigned a learning course on the LMS. Learners browse the textbooks, answer quizzes, and submit assignment files in the course on the LMS. All of these activities are then recorded as data on LMS. Goda et al. (2015) investigated learning patterns based on learning logs in computer-assisted language learning settings. They compared with performances between students in each learning patterns, and found that the students of procrastination type were lower scores than some other types. Procrastinating learners attract attention at any age. In asynchronous online classes, there are no restrictions on learning time and students can learn at any time, making it easier for them to procrastinate. In an asynchronous online class, learners' activities are recorded in the LMS, and based on these records, learners with procrastinating behavior are predicted in advance. The prediction results should be useful for instructing learners who are likely to procrastinate in advance.

We provide information literacy classes with about 1800 students. In this class, learners do not meet in the classroom, but browse the textbook on the LMS and work on the exercise problems using computers or tablets by themselves. The course textbook contains explanations of the learning contents and weekly assignments. The LMS (Moodle) records not only the record of the learner's browsing the textbook, but also the id number of the chapter attribute (chapterid), which corresponds to the page of the textbook. Considering that learners change their behavior between the current week and the week before due to procrastination behavior, Kubota (2021) constructed characteristic vectors in each learner using the records of the pages of the textbook accessed by the learners during the week they were studying (the w week) and the week before that (the w-1 week). Using the characteristic vectors in each learner, we classified learners into two classes, procrastination behavior or not, by the machine learning method.

In this approach, learners with procrastination behavior were predicted using two weeks of LMS data. For example, if the prediction was to be made in the 6th week, then the LMS data from the 5th and 6th weeks will be used for the prediction. The training data consisted of two weeks of LMS data of learners who failed to submit the assignment by the deadline one week before the week of prediction. For example, if the prediction was to be made in the 6th week, the training data will consist of the LMS data from the 4th and 5th weeks of the learners. This approach was evaluated using the LMS data from our asynchronous online class in the first semester of 2020. The maximum number of learners who failed to submit their assignments by the deadline was 36. The number of positives, which means learners who failed to submit their assignments by the deadline, was small compared to the number of negatives. Since the training data was imbalanced, we sampled from majority class in order to balance dataset between positives and negatives. Using the sampled dataset, we calculated the prediction accuracy over 15 weeks. The accuracy was about 50% and the characteristic vectors and prediction methods should be improved more.

Keywords:

Learning Analytics, Page Transition, Procrastination Behavior.