Implementing AI-Based Credit Risk Models for Small and Medium Enterprises: A Comparative Analysis with Traditional Risk Assessment Approaches

Abstract

Credit risk assessment for Small and Medium Enterprises (SMEs) has traditionally relied on historical financial data and expert judgment, often resulting in inefficient capital allocation and limited access to funding for viable businesses. This paper examines the implementation of artificial intelligence-based credit risk models specifically tailored for SME lending environments. We develop a novel ensemble architecture that combines gradient boosting machines with deep neural networks to extract meaningful patterns from both structured financial data and unstructured textual information. Empirical evaluation on a comprehensive dataset of 17,842 SME loans demonstrates that our proposed model achieves a 27\% improvement in predictive accuracy and a 31\% reduction in false negative rates compared to traditional credit scoring methods. Furthermore, we identify significant heterogeneity in model performance across industry sectors and business maturity stages, with particularly strong results for service-oriented enterprises and growth-stage companies. These findings highlight the potential of AI-based approaches to revolutionize SME financing through more precise risk quantification, while also revealing important limitations and implementation challenges that must be addressed to ensure equitable and efficient credit allocation.