인공지능 딥 러닝 기법을 이용한 영어 시험의 성적 예측과 활용

Prediction of English test results using artificial neural network.

This study researches the possibility of applying a deep learning-based English assessment score prediction model in an assessment setting to predict scores of unperformed English assessments. A reading difficulty index and a corpus index were calculated as attributes connected to each test type for each English test type. A deep learning score prediction model was run to learn from the test. The model's validity was determined by comparing the predicted model's performance on a new trial with the scores obtained from the actual test. The results show a high level of prediction accuracy, and out of the 10,380 participants' scores from the TOEIC English assessment taken as a part of a university liberal arts class over three years, for 782 test-takers, there was only a 0.07- point difference in the average scores across 3 test types. Furthermore, there was only a 1.5 and 1.3 point difference between the top 30% and bottom 30% groups. It was also confirmed that the results were valid even when the individual test takers were replaced with a different group. It is expected that this prediction model of English test results can be used as an additional indicator for developing an English test. This study proposes that the valid result predictions of unperformed English assessments using the deep learning-based English assessment prediction model can be used as a new supporting indicator.