Student learning performance prediction based on online behavior: an empirical study during the COVID-19 pandemic

Abstract

In the context of the COVID-19 global pandemic, highly intense and frequent online teaching has leapt to be one of the dominant learning patterns and become an ordinary situation in university teaching practices. In recent years, progress in feature engineering and machine learning has made it possible for more effective educational data mining, which in turn has enhanced the performance of intelligent learning models. However, the potential impact of increasing and varying features on online instruction in this new situation makes it unclear whether the existing related findings and results are practical for teachers. In this article, we use various state-of-the-art machine learning techniques to predict students’ performance. Based on the validation of the rationality of the built models, the importance of features under different feature selection techniques are calculated separately for the datasets of two groups and compared with the features before and at the beginning of the pandemic. The results show that in the current new state of highly intense online learning, without considering student information such as demographic information, campus attributes (administrative class and teaching class) and learning behavior (completion of online learning tasks and stage tests) these dynamic features are more likely to discriminate students’ academic performances, which deserves more attention than demographics for teachers in the guidance of students’ learning. In addition, it is suggested that further improvements and refinements should be made to the existing features, such as classifying features more precisely and expanding in these feature categories, and taking into account the statistics about students’ in-class performances as well as their subjective understanding of what they have learned. Our findings are in line with the new situation under the pandemic and provide more implications to teachers’ teaching guidance.

Keywords

• Educational data mining
• Machine learning
• Students’ performance
• Online learning