Comparative Study of Traditional and Machine Learning Models for SME Credit Risk Assessment

DOI

10.2991/978-94-6239-774-3_34How to use a DOI?

Keywords

SME credit risk; small business lending; logistic regression; support vector machine; random forest; XGBoost; SBA 7(a) loans; ROC-AUC; precision-recall analysis; feature importance

Abstract

Small and medium-sized enterprise (SME) credit risk assessment is an essential challenge for financial institutions because SME borrowers are characterized by high economic relevance, information opacity, diverse operating histories, and sensitivity to loan terms. This study presents an empirical comparison of traditional and machine learning models for SME credit risk assessment using official U.S. Small Business Administration 7(a) Freedom of Information Act (FOIA) open data. Loans approved in FY2010-FY2019 are used to develop the models, while loans approved from FY2020 through the current release are used as an out-of-time test sample. Charged-off loans are treated as default observations, and paid-in-full loans are treated as non-default observations. To minimize information leakage, variables that reveal post-origination outcomes are excluded. Logistic regression is used as the traditional benchmark, and its performance is compared with three challenger models: linear support vector machine, random forest, and XGBoost. The empirical results show that XGBoost performs best on the out-of-time test set in terms of accuracy, precision, recall, F1-score, ROC-AUC, PR-AUC, and Brier score (accuracy 0.8981, precision 0.4416, recall 0.8199, F1-score 0.5741, ROC-AUC 0.9193, PR-AUC 0.5838, and Brier score 0.0754). Feature analysis indicates that maturity, interest-rate structure, secondary-market sale status, SBA subprogram, lender geography, revolving status, guarantee amount, and business age are important risk signals. The findings confirm that machine learning can improve SME default identification under a temporal validation design while preserving financially meaningful model interpretation. Tree-based learning provides stronger discrimination and recall and captures nonlinear interaction effects more effectively, whereas logistic regression remains a useful transparent baseline.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Siqi Huang
PY  - 2026
DA  - 2026/09/11
TI  - Comparative Study of Traditional and Machine Learning Models for SME Credit Risk Assessment
BT  - Proceedings of the 2026 5th International Conference on Mathematical Statistics and Economic Analysis (MSEA 2026
PB  - Atlantis Press
SP  - 358
EP  - 371
SN  - 2352-5428
UR  - https://doi.org/10.2991/978-94-6239-774-3_34
DO  - 10.2991/978-94-6239-774-3_34
ID  - Huang2026
ER  -