Predicting students' performance of an offline course from their online behaviors

Prediction of students' academic performance is a worthwhile task for educational institutions, as they may provide necessary help to at-risk students as early as possible. Previous studies mainly focused on predicting students' success or failure of online courses. We show that it is possible to predict students' performance of offline courses from their access records on general websites. Feature set of our prediction model includes the number of records on various categories of websites, scores of another course delivered in last semester, and the amount of time spent on online videos. Experiments demonstrate that the proposed model can predict where a student can pass Data Structure course early in the midterm, with a specificity of above 65% and a sensitivity of nearly 90%.