Beyond First-Year: Predicting Mid-Progress Dropout at an HBCU

Session Abstract: Most dropout prediction focuses on first-year students, yet the highest attrition at an HBCU occurs in the mid-progress window (30–95 credit hours), where students are beyond initial transition supports but still far from graduation. Using 2019–2023 enrollment and financial data, we compared Logistic Regression, Random Forest, Gradient Boosting, XGBoost, CatBoost, and LightGBM with a multi-term dropout definition. F1 scores exceeded 0.80, with CatBoost achieving recall above 90%. SHAP interpretability showed that academic momentum (credit hours and GPA trajectory), loan burden, online enrollment, and age/dependency status drive risk. We translate these predictors into actionable interventions - degree audits, aid counseling, and targeted advising for online and older students, offering a transferable playbook to strengthen mid-progress retention at HBCUs and similar institutions.

Keywords: HBCU; Retention; Dropout Prediction; Machine Learning; Ensemble Methods