MACHINE LEARNING ENHANCED CREDIT RISK ASSESSMENT MODELS FOR COMMUNITY AND REGIONAL BANKS: IMPROVING LENDING DECISIONS THROUGH ALTERNATIVE DATA AND PREDICTIVE ANALYTICS

Abstract

Community and regional banks play a critical role in providing credit to small businesses, farmers, and individuals in markets that larger national banks often underserve. However, their credit risk assessment practices have traditionally relied on conventional bureau scores and limited financial data, which can lead to suboptimal lending decisions for borrowers with thin credit files or unconventional financial profiles. This paper proposes a machine learning enhanced credit risk assessment framework specifically designed for community and regional banks, integrating alternative data sources with predictive analytics to improve lending decisions. The framework combines traditional credit bureau data with alternative signals including transaction history, utility and rental payment records, business cash flow patterns, and digital footprint indicators. A layered modelling approach uses gradient boosting for primary risk scoring, neural networks for sequential behavioural analysis, and explainability tools to support regulatory and operational requirements. Empirical evaluation was conducted on a simulated lending portfolio reflecting community bank characteristics, with outcomes measured against conventional credit scoring baselines. Results show that the proposed framework improved discrimination performance from an AUROC of 0.74 to 0.86, reduced the false rejection rate for creditworthy applicants by 31 percent, and increased approval rates for thin-file borrowers by 28 percent while maintaining default rates at acceptable levels. The findings demonstrate that machine learning combined with alternative data offers community and regional banks a credible path to better lending decisions and expanded financial access.