Dropout Prediction by Interpretable Machine Learning Model Towards Preventing Student Dropout

Abstract

In the education industry, the needs of online learning are significantly increasing. However, the web-based courses demonstrate higher dropout rates than traditional education courses. As a result, engaging students with data analysis is getting more crucial especially for distance learning. In this study, we analyze data on the daily learning status of students in order to predict the student's dropout in online schools. Specifically, we trained a dropout prediction machine leaning model with 1) Basic attributes of students, 2) Progress of learning materials, and 3) Slack conversation data between students and teachers. The experimental results show that the accuracy rate of the machine learning model has reached 96.4%. As a result, the model was able to predict 78% of the students who actually dropped out of school. We also looked into feature importance by SHAP value to gain ML model interpretability.