Title: Using tree-based models to predict credit risk

Authors: Nathan Coates; Robert Nydick; D.K. Malhotra

Addresses: Villanova University, 800 E. Lancaster Ave., Villanova, PA 19085, USA ' Villanova University, 800 E. Lancaster Ave., Villanova, PA 19085, USA ' Thomas Jefferson University, 4201 Henry Avenue, Philadelphia, PA 19144, USA

Abstract: Despite an increase in consumer bankruptcies, the consumer loan industry is increasingly competitive. Financial organisations may find that well-allocated credits are one of the most lucrative sources of income. However, a high degree of risk is associated with this type of banking activity because many incorrect judgements might force the lending institution into bankruptcy. The main goal of credit risk evaluation research is to develop classification rules that properly classify bank clients as either good credit or bad credit loan applicants. This study shows how to use tree-based algorithms, such as decision trees, random forests trees, boosted trees, and XGBoost, to lower the risk of bad loans and find the traits that can help differentiate between a good loan and a bad loan. This will allow loan officers to improve their scoring models by giving those traits more weight when deciding whether to extend loans to borrowers. Lending institutions can protect themselves from legal or regulatory problems by explaining the factors that led them to decide against lending to a potential borrower.

Keywords: consumer loans; credit risk; decision trees; bootstrap trees; boosted trees.

DOI: 10.1504/IJBISE.2024.139147

International Journal of Business Intelligence and Systems Engineering, 2024 Vol.2 No.1, pp.23 - 42

Received: 29 Aug 2022
Accepted: 09 Nov 2022
Published online: 15 Jun 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article