Leveraging Ensemble Machine Learning for Credit Risk Assessment in Underserved U.S. Small Businesses

This study investigates the role of ensemble machine learning techniques in improving credit risk assessment for underserved small businesses in the United States. Conventional credit evaluation systems rely heavily on formal financial documentation and collateral requirements, excluding a large proportion of early-stage, minority-owned, or cash-based enterprises. Drawing on the uploaded document's empirical modeling structure, the study integrates gradient boosting frameworks, random forest classifiers, and macro-behavioral feature engineering to construct a multidimensional risk assessment model. The methodology incorporates structured preprocessing, cross-validated ensemble training, and interpretability analysis using SHAP values. Results show that boosting-based models consistently outperform traditional approaches, achieving stronger precision, recall, and AUC scores while capturing nuanced behavioral and macroeconomic interactions. The findings highlight the pot ential of ensemble learning to expand access to credit, reduce misclassification bias, and support inclusive economic development. The study concludes with recommendations for lenders, policymakers, and researchers regarding the deployment of ensemble analytics and the future integration of fairness-aware modeling.