[Paper Review] Data-Driven Exploration of Factors Affecting Federal Student Loan Repayment

Bin Luo, Qi Zhang+1 more|arXiv (Cornell University)|May 3, 2018

Higher Education Research Studies14 references3 citations

TL;DR

This study uses machine learning and statistical models on the U.S. College Scorecard dataset to identify key factors influencing federal student loan repayment rates. It finds that student family income, Pell Grant receipt, and early withdrawal rates are the most predictive factors, with Random Forest outperforming linear and elastic-net models in prediction accuracy (RMSE = 0.0153).

ABSTRACT

Student loans occupy a significant portion of the federal budget, as well as, the largest financial burden in terms of debt for graduates. This paper explores data-driven approaches towards understanding the repayment of such loans. Using statistical and machine learning models on the College Scorecard Data, this research focuses on extracting and identifying key factors affecting the repayment of a student loan. The specific factors can be used to develop models which provide predictive capability towards repayment rate, detect irregularities/non-repayment, and help understand the intricacies of student loans.

Motivation & Objective

• To identify the most influential factors affecting federal student loan repayment rates using data-driven methods.
• To compare the predictive performance of machine learning models (Random Forest, Elastic-Net) against traditional linear models.
• To examine the role of student demographics, institutional characteristics, and financial aid in shaping repayment outcomes.
• To support policy and lending decisions by uncovering latent relationships in federal loan data.
• To improve predictive modeling of repayment behavior using feature importance and cross-validation techniques.

Proposed method

• Applied Principal Component Analysis (PCA) to reduce dimensionality and identify latent factors in the College Scorecard dataset.
• Used Elastic-Net regression with L1 and L2 regularization to perform feature selection and improve model generalization.
• Implemented Random Forest Regression with 20 decision trees and 10-fold cross-validation to predict repayment rates.
• Calculated feature importance using permutation-based variance reduction to rank predictors.
• Evaluated model performance using Root Mean Squared Error (RMSE) and compared across models.
• Conducted correlation analysis and feature clustering to validate model findings and explore inter-variable relationships.

Experimental results

Research questions

• RQ1 Which student, institutional, and financial aid factors most strongly predict federal student loan repayment rates?
• RQ2 How do machine learning models like Random Forest and Elastic-Net compare in predicting loan repayment compared to traditional linear models?
• RQ3 To what extent do family income, Pell Grant status, and early withdrawal rates influence repayment outcomes?
• RQ4 Are there consistent feature importance patterns across multiple modeling techniques?
• RQ5 Can unsupervised or penalized feature selection methods improve model accuracy and interpretability?

Key findings

• The Random Forest model achieved the best performance with an RMSE of 0.0153, outperforming both linear regression and Elastic-Net models.
• Family income (FAMINC) was the most important predictor, indicating that higher family income correlates strongly with lower default risk.
• Pell Grant receipt (PCTPELL) was the second most important feature, suggesting that financial aid significantly improves repayment likelihood.
• Early withdrawal rates (WDRAW_ORIG_YR2_RT) and enrollment shares of Black and White students (UGDS_BLACK, UGDS_WHITENH) were also top predictors.
• The Elastic-Net model identified 'Aid' and 'School' categories as relevant, which were not highlighted by correlation analysis alone.
• Feature importance was consistent across models, with 'Student' and 'Completion' categories dominating predictive power.