Predicting quitting behavior of students incorporating contents of exercises in online education services

Abstract

If the quitting behavior of a student can be accurately predicted in online education services, it is possible to gain insight into the state of the student in real-time and perform interventions to guide the student in a direction not to quitting. In addition, understanding the tendency and intention of the behavior leads to the development of useful teaching materials for students. In this paper, we attempt to predict the quitting behavior of students incorporating the contents of exercises. The proposed method predicts the probability of whether the student will quit or not in a session using the log data stored in the service. We extract not only the features related to the students' actions but also the features related to the contents of exercises. The features related to the contents of exercises are extracted from multiple viewpoints such as the stay time of the students who answer each teaching material and the text included in the exercises. In the experiment, we use actual log data of students in the programming learning service Aidemy and verify the effectiveness of incorporating contents of exercises in the prediction of the quitting behavior.