Statistical and machine learning models for predicting university dropout and scholarship impact

Area of Science:

• Higher Education Research
• Educational Data Mining
• Student Retention Studies

Background:

• Student dropout is a persistent challenge in higher education.
• Understanding dropout predictors is crucial for developing effective intervention strategies.
• Previous research often lacks a causal analysis of financial aid's impact on retention.

Purpose of the Study:

• To identify key socioeconomic and academic features predicting student dropout risk.
• To causally assess the impact of scholarship awards on student retention.
• To compare the performance of various machine learning models in predicting dropout.

Main Methods:

• Analysis of 4,424 student records from a Portuguese higher education institution.
• Definition of dropout includes field and institution changes.
• Propensity score matching used to estimate the causal effect of scholarships.
• Evaluated classifiers: Lasso regression, generalized additive model (GAM), random forest, XGBoost, and neural networks.

Main Results:

• XGBoost model achieved the highest F1-score (0.904).
• Key predictors include second-semester grades, credited units, tuition payment status, debt, scholarship holding, and enrollment age.
• GAM analysis indicated scholarships reduce dropout odds by ~40% (22.2% probability reduction).

Conclusions:

• Academic performance and financial status are significant factors in student dropout.
• Scholarship programs demonstrably improve student retention.
• Machine learning models, particularly XGBoost, offer powerful tools for predicting dropout risk.