Small Business Administration Guaranteed Loan Default Detection Model with Optimized Boosting Methods
Open AccessSmall Business Administration (SBA) uses credit score models to control the risk of loans. Loan default detection has become increasingly important as the high average charge-off increases from 2011 to 2020. This praxis presents an optimized boosting machine learning model to estimate the probability of default of the SBA guaranteed loans during COVID-19 crisis, using the SBA 7(a) loan data from 2010 to 2021. Statistical analysis methods were used to identify the most essential factors of borrower, lender, as well as loan characteristics that were relevant to SBA loan default. The ten most important factors based on the machine learning model were identified and include borrower region, business type, industry type, franchise status, loan delivery type, loan term, loan approval amount, and revolver status. A baseline logistic regression was built based on the factors selected from the study variables. Four advanced machine learning models with bagging or boosting methods were fitted to compare with the baseline logistic regression model. The optimized LightGBM model, which had the highest recall of 0.95 on the test dataset, was selected as the final output of the praxis. The SHAP value analysis was applied to interpret the final output model globally and locally. The goal of this praxis is to improve the current risk management of SBA loan approval process with statistical analysis and machine learning models, which help to estimate the probability of default in future SBA 7(a) loans during COVID-19 crisis.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.