2026 ASEE Annual Conference & Exposition

Reducing attrition rates among STEM-enrolled college students is critically important to universities. Improved retention and higher graduation rates lead to better academic outcomes for both students and their respective institutions. This WIP paper examines whether social dynamics among co-enrolled students in STEM courses, combined with data encapsulating student grades, self-efficacy, and academic preparedness, can predict the risk of failure or withdrawal for engineering and non-engineering majors. Key performance metrics are collected throughout the semester. These include grade data over tests, quizzes, and homework, along with responses to post-exam surveys. The surveys gather benchmark performance data and assess peer interaction patterns, including motivation levels, grade expectations, number of study peers, study time together, and study locations. Both survey and grade data are used to train statistical learning models designed to classify students at risk. Improved model fidelity when incorporating social variables would indicate a promising direction for predicting educational outcomes. In large classrooms, struggling students often go undetected by instructors; high-performing models would enable instructors with the ability to automatically identify at-risk students early. This would allow instructors to direct support resources toward at-risk students more effectively and at their discretion. Improved support in engineering prerequisites like introductory physics and mathematics, where classroom sizes often exceed 50 students, can lower overall attrition and improve graduation rates. Thus, improving detectability of at-risk students in the classroom for engineering majors would increase the rate of engineering graduation, and thereby increase the future engineering workforce.