MACHINE PERCEPTION AND ROBOTICS GROUP

Dept. of Robotics Science and Technology,
Chubu University

With the digitization of educational materials, there is growing anticipation for predicting grade performance using learning log data. Previous studies have attempted to predict performance by inputting histogram features of the number of digital material operations into machine learning models. However, these approaches do not consider temporal sequences, making it difficult to reflect behavioral patterns in the performance predictions. To address this issue, we propose maintaining the time series using fastText, which embeds learning behaviors as features. Additionally, we employ differential pattern mining to detect behavior patterns that exhibit significant differences and then apply weighting to these patterns in fastText. Evaluation experiments show that our proposed method improves performance prediction accuracy compared to conventional methods and that weighting behavioral patterns proves effective.