Technical gazette, Vol. 30 No. 2, 2023.
Original scientific paper
https://doi.org/10.17559/TV-20221220085239
An Ensemble Learning Based Strategy for Customer subdivision and Credit Risk Characterization
Shuaiqi Liu
; School of Economics & Management, University of Science and Technology Beijing, No.30 Xueyuan Road, Beijing 100083, China
Guiying Wei
; School of Economics & Management, University of Science and Technology Beijing, No.30 Xueyuan Road, Beijing 100083, China
Sen Wu
; School of Economics & Management, University of Science and Technology Beijing, No.30 Xueyuan Road, Beijing 100083, China
Yiyuan Sun
; School of Economics & Management, University of Science and Technology Beijing, No.30 Xueyuan Road, Beijing 100083, China
Abstract
Credit customer subdivisions and borrower characteristics are essential tools for banks and lending companies to evaluate credit risk and make profits. This study proposes a Multi-Level Default Risk Rating (MLDRR) strategy based on a heterogeneous ensemble learning method. Further, a novel Eight Subdivisions Model (ESM) of credit customers is constructed. Through the model, the credit customers are subdivided into eight important categories, such as the defaulting customers that are easily missed, the customers with the highest risk, customers with the potential risk, and target customers, etc. Moreover, we describe the risk characteristics of the customer subdivisions and find deficiencies in the agency's existing risk ratings. Finally, the explored strategies are validated on one hundred thousand real credit data, demonstrating the effectiveness of ESM. Compared with the traditional customer segmentation and characteristics research, this paper develops a new credit customer segmentation method based on the perspective of default risk and describes the risk characteristics more comprehensively through eight customer subdivisions.
Keywords
credit customer subdivision; credit risk rating; customer characterization; ensemble learning
Hrčak ID:
294339
URI
Publication date:
26.2.2023.
Visits: 974 *