Machine Learning Algorithms in Small Business Credit Underwriting

Abstract

Small business credit underwriting has undergone a massive transition in the last two decades with the rise of machine learning algorithms that enhance the accuracy and efficiency of credit risk assessments. In this paper, I study how machine learning models, including XGBoost, Random Forest, and k-Nearest Neighbors, can be applied to improve upon traditional credit scoring methods. By analyzing a dataset from the U.S. Small Business Administration, I evaluate how these models perform in predicting loan defaults, uncovering complex patterns in financial and behavioral data. In conclusion, the results demonstrate that machine learning models, particularly XGBoost, significantly outperform traditional methods, achieving higher accuracy in predicting credit risk. However, challenges such as data privacy concerns, algorithmic bias, and model transparency must be addressed to ensure ethical and reliable use in underwriting. The findings suggest that with responsible implementation, machine learning can optimize small business lending decisions and promote financial inclusion.