Case Study of Proficiency-Based Class Composition Using Machine Learning

This paper reports a case study of creating a prediction model of English placement test scores for proficiency-based English language classes for incoming first-year students. Since the English placement test could not be administered in FY2020 due to the COVID-19 pandemic, we used data from the past five years to create a random forest regression model was created using data from the past five years. Since the variables that could be used were limited to fixed information at the time of admission, the number of variables was minor, such as the type of entrance examination and high school types. Although the performance metrics of the model were not perfect, it was judged to be practical for using the prediction results, namely, class placement. In the following year, 2021, the English placement test could be administered, so the prediction model created in 2020 was used to verify the prediction results and actual performance. As a result, it was confirmed that the performance evaluation index tended to decrease. However, the performance was judged to be within the practical range, and the generalization performance of the forecasting model was confirmed. In addition, when the importance of the variables used in the prediction model was checked, the importance of the recommended entrance examination system was high. This suggests that the difference in English proficiency at the time of entrance is mainly due to the entrance examination system and the future that devise a countermeasure is necessary such as remedial education.