Learning from leavers: Early academic predictors of engineering attrition via machine learning

• In Event: AERA Roundtable Session Sunday 11:45 am - Gold 1
  In Roundtable Session: Roundtable 6

Abstract

Despite national efforts to increase engineering retention, over half of students who begin in engineering leave before completing a degree. This study uses machine learning to identify early predictors of attrition among undergraduate engineering students at a large public university. Using longitudinal data from 2012 to 2023, we modeled departure between academic Terms 2 and 5 using decision trees, random forests, and conditional inference trees. Results show that while demographic factors had limited predictive power, early college GPA was consistently a strong predictor of persistence. High school GPA declined in importance over time, and critical course grades were only moderately predictive. Findings highlight the value of early academic monitoring and support during students’ first year to improve retention in engineering.

Authors

• Lydia Ross, Arizona State University

• Amanda J. Davis Simpfenderfer, College of William & Mary

• Travis Terpkosh, Arizona State University

• Junzhao Wang, Arizona State University