Building Course-Specific Regression-based Models to Identify At-risk Students

The first step in helping students who may fail a course is to identify them as early in the semester as possible. Predictive modeling techniques can be used to create an early warning system which predicts students’ success in courses and informs both the instructor and the students of their performance. One common problem with existing early warning systems is that they typically employ a general model that cannot address the complexity of all courses. In this study, we built three models to identify at-risk students in a specific large first-year engineering course at three important times of the semester according to the academic calendar. Then the models were optimized for identifying at-risk students. The models were able to identify 79% of at-risk students at week 2, 90% at week 4, and 98% at week 9. This high accuracy illustrates the value of creating course specific prediction models instead of generic ones.