Integrating Students' Behavioral Signals and Academic Profiles in Early Warning System

In this paper, we investigated how students' behavioral signals and incoming profiles can be integrated to describe and predict student success in a higher education's STEM course. The results include three major findings. First, we found behavioral signals like the number of correct responses to in-class questions, the number of confusing slides, and the number of viewed slides and videos are stable predictors of student success across different periods of a semester. Second, from the mixed-effect modeling results, we could identify significant gender gaps between mid-level incoming GPA student groups. We also showed some possible course advising scenarios based on the interaction between student behaviors and incoming profile factors. Third, using both behavioral signals and incoming profiles, our weekly forecast model on student success achieved a 72% prediction accuracy. We believe these findings can set the stage for subsequent early warning system studies that use different types of student data. Further investigations on the causal relationships for suggested results and developing other novel predictive features for student success would be beneficial for designing a better early warning system.