Identifying Student Behaviors Early in the Term for Improving Online Course Performance

In this study we investigate the correlation between student behavior and performance in an online course. We introduce highlevel behavior features derived from the course syllabus and sequential patterns. We propose a random forest algorithm with cross-validation to find correlation between behavior features and student performance. Considering a course with 10 periods, our empirical results indicate that our models can reach 70% accuracy from behavior features in the first period and 90% from features in the first 5 periods. We identify both important individual behaviors and behavior combinations; our results indicate starting to study earlier is important in both types of behavior.