Hasan, Md Majedul and Rahim, Muhammad and Ifti, Naim Rahman and Nafees, Muhammad Shayan (2026) Beyond Predicting Failure: A Cost-Sensitive, Explainable AI Framework for Identifying Actionable Recovery Pathways in the U.S. Small Businesses. American Journal of Economics and Business Management, 9 (10). ISSN 2576-5973

Text AJEBM_Beyond+Predicting+Failure.pdf Download (862kB)

Abstract

Small companies are a critical component of the U.S. economy, providing jobs, innovations and economic growth. However, many small businesses suffer financial difficulties because they lack access to capital, are not operating efficiently and due to market fluctuations, they default on their loans and businesses fail to succeed. The traditional predictive models are limited in interpretability for the decision-makers and mainly aim to predict failure of the business without taking into account the costs of misclassification. This study introduces a new framework, a Cost-Sensitive, Explainable Artificial Intelligence (AI) Framework, which goes beyond the prediction of failure and also determines actionable pathways for recovery for U.S. small businesses. The proposed framework involves the use of the National Small Business Administration (SBA) Loan Dataset (SBAnational.csv) which contains 899,164 loan records and 27 business and loan related variables to build predictive models for financial distress. The data is extensively preprocessed and engineered, after which various machine learning techniques, such as Logistic Regression, Decision Tree, Random Forest, XGBoost, LightGBM and CatBoost are tested under a cost-sensitive learning approach to reduce the financial harm of non-symmetric misclassification errors. The use of Explainable AI (XAI) techniques like SHAP and LIME allows for the extraction of insights from the models, thereby helping to identify factors that are crucial for business distress, including loan characteristics, business size, industry classification, and financing conditions. This is a novel approach of translating AI-generated explanations into actionable recovery paths, which enables informed managerial decision making, such as recommendations to enhance liquidity management, financing structures and operational risk management, and long-term business resilience. The accuracy, precision, recall, F1 score, ROC-AUC, Balanced accuracy, Matthews correlation coefficient, and cost-sensitive evaluation metrics are used to evaluate the model performance. The proposed framework will add to the domains of Business Analytics and Explainable AI by combining the three pillars of predictive accuracy, interpretability, and practical decision support in one. The results will help small business owners, financial institutions, policy makers, and the U.S. Small Business Administration to proactively detect financial distress and take evidence-based recovery measures to enhance the sustainability and resilience of small businesses.

Item Type:	Article
Uncontrolled Keywords:	Cost-Sensitive Machine Learning, Explainable Artificial Intelligence (XAI), Financial Distress Prediction, U.S. Small Businesses, Actionable Recovery Pathways and Business Analytics
Subjects:	H Social Sciences
Depositing User:	admin eprints
Date Deposited:	22 Oct 2026 02:48
Last Modified:	22 Oct 2026 02:48
URI:	http://eprints.umsida.ac.id/id/eprint/17263

Actions (login required)

View Item