Comparing StepAIC and LASSO Regression Models to Predict Student Retention

Session Abstract: Student retention is a vital measure of institutional performance and student success, especially in community colleges serving diverse and non-traditional populations. This study examined data on 7,419 students enrolled at Kaskaskia College between 2020 and 2025 to evaluate fall-to-fall retention using demographic, academic, and financial predictors. Two modeling approaches were compared: stepwise logistic regression with Akaike Information Criterion (StepAIC) and penalized logistic regression using LASSO. Both achieved similar predictive performance (accuracy 68.4%; AUC ~0.74). StepAIC yielded interpretable factor-level coefficients, while LASSO retained more predictors with smaller coefficients. A DeLong test showed no significant difference between models. Together, these methods offer complementary insights to inform interventions, early-alert systems, and predictive dashboards that enhance student persistence and institutional effectiveness.

Keywords: Student Retention, Predictive Analytics, Community College, Logistic Regression, Model Selection