Automatic assessment of student understanding for estimateing test score

Prediction of the level of student understanding to learning is one of crucial issues in education. The level of understating is usually estimated by simulation test scores from the assumption that there is a strong relationship between the level of understanding and a simulation test score. Meanwhile, instructors or teachers often assess student understanding by observing their learning status and use the assessment scores to give the students feedback. However, since the assessment scores to student understanding are obtained subjectively, the distribution of the understanding scores tends to be biased toward extremely high scores; consequently the assumption of the relationship to simulation test scores would not be held. In this paper, we propose a method to automatically rate student understanding scores based on instructors' observation reports using supervised learning method.In order to evaluate our proposed method, we conducted experiments on real data provided by a cram school. In the experiments, we estimate simulation test scores on a students' 60-point math exam using student understanding scores rated by our proposed method and those manually rated by instructors for comparison. The results of experiments show that the MAE between the understanding scores rated by our proposed method and those manually rated by the instructors were 7.81 and 8.73, respectively, which indicates that our proposed model significantly improve rating of the level of student understanding to learning in mathematics.