Predicting student performance (PSP) is of great use from an educational perspective, especially for the at-risk studentswho need timely support to complete their study. Previous PSP studies have been mainly based on data fromquestionnaires and specific learning systems. Such data sources have some innate shortcomings. Instead, we used anovel data source, the massive students’ campus card usage records, to predict at-risk students. This method has twoadvantages: convenience in data collection and ability to predict students’ overall academic performance. However, as theoriginal data are complex, large in scale and with a lot of noise, it is challenging to extract proper features from them. Weadopted a four-step procedure for data preprocessing and the Naive Bayes model for performance prediction. Experimentsshowed that the proposed prediction model could identify about 70% of the at-risk students. Some features of the at-riskstudents were also discovered, which might help student counselors and educational researchers better understand therelationship between students’ consumption behaviors and their academic performance.
© 2001-2024 Fundación Dialnet · Todos los derechos reservados