Using Bayesian networks to perform reject inference. (15th December 2019)

• Record Type: Journal Article
• Authors: Anderson, Billie
• Is Part Of: Expert systems with applications. Volume 137(2019)
• Page Start: 349
• Page End: 356
• Publication Date: 2019-12-15
• Subjects: Credit scoring -- Reject inference -- Bayesian networks -- Predictive accuracy; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2019.07.011
• Languages: English
• ISSNs: 0957-4174

Abstract

Highlights: Reject inference substantially increases the quality of a credit scorecard. Bayesian networks show promise as reject inference method since no functional form assumed. Bayesian networks outperform standard reject inference methods. Abstract: Credit scoring is an automatic credit assessment tool that has been used by different types of financial institutions for years. When a financial institution wants to create a credit scoring model for all applicants, the institution only has the known good/bad loan outcome for the accepted applicants; this causes an inherent bias in the model. Reject inference is the process of inferring a good/bad loan outcome to the applicants that were rejected for a loan so that the updated credit scoring model will be representative of all loan applicants, accepted and rejected. This paper presents an empirical reject inference technique using a Bayesian network. The proposed method has an advantage over traditional reject inference methods since there is no functional form that will be estimated with the accepted applicants' data and extrapolated to the rejected applicants to infer their good/bad loan outcome status.