Credit Default Risk and Client Segmentation in Agricultural Lending

Abstract

This study develops a predictive credit scoring model for the National Agricultural Bank (BNA) to improve the as-sessment and management of default risk among agricultural borrowers, who are highly exposed to financial vulner-ability and environmental uncertainty. Using the CRISP-DM methodology, the research includes data preprocessing, exploratory analysis, and model comparison between Quadratic Discriminant Analysis, Random Forest, and XGBoost, followed by out-of-time validation. The proposed model integrates agriculture-specific financial indicators with ecological variables such as climate and regional temperature variability, which significantly influence rural credit risk. Findings reveal that XGBoost achieves the highest predictive accuracy and robustness, particularly in de-tecting high-risk clients. By combining financial and environmental dimensions, this approach strengthens BNA’s credit risk management and supports sustainable lending practices in the agricultural sector. Additionally, the model enables strategic marketing applications by segmenting borrowers into distinct risk-based personas, facilitating the design of tailored financial products and communication strategies. These insights contribute to improved customer engagement and more inclusive agricultural finance, aligning with the bank’s objectives of resilience and sustainabil-ity.