Predictive modeling and cohort data analytics for student success and retention

Abstract

This study presents a data-driven analysis of academic performance, demographic disparities, and predictive modeling among more than 23,000 first-time freshmen at a US public University. We examine multiple factors influencing student outcomes, including GPA, credit accumulation, unit workload, Pell Grant eligibility, minority status, and parent education levels. Our analysis reveals several statistically significant disparities: non-minority students earn more units than minority students in their first two years, and Pell-eligible students accumulate fewer credits than their non-eligible peers. First-generation college students also exhibit lower credit accumulation compared to peers. GPA distributions show that minority students have a lower average GPA compared to non-minority students, with broader variation. Clustering analysis identifies three distinct academic engagement profiles based on GPA and unit load, highlighting heterogeneous performance patterns and the need for differentiated support. We develop and tune predictive models to forecast sophomore credit accumulation and GPA, achieving strong performance using deep learning. These models enable proactive risk identification and support strategic interventions. Our findings set the stage for actionable insights for institutional decision-makers aiming to enhance student retention, success, and academic momentum.

Keywords: Data analytics; Education; Higher education; Retention; Student success.