Work

Advancing financial risk management: A transparent framework for effective fraud detection

Year: 2025

Type: article

Abstract: Robust financial fraud detection is crucial for protecting assets and maintaining financial system integrity. Traditional models lack flexibility, while machine learning models are often complex and difficult to interpret. We propose an XGB-GP framework that combines Extreme Gradient Boosting (XGB) and Genetic Programming (GP) to create interpretable models, enhancing fraud detection. Our framework highlights the effectiveness of the financial indicator “Total Liabilities/Operating Costs” and outperforms traditional and machine learning models in detecting fraud, as demonstrated through analysis of data from the CSMAR database of Chinese publicly listed companies. • We highlight the key role of different indicators in detecting financial fraud. • We identify a novel financial indicator to improve fraud detection accuracy. • We apply rigorous preprocessing to reduce false-negative impacts in detection. • We present a user-friendly tool that addresses common black-box model issues.

Source: Finance research letters

DOI: 10.1016/j.frl.2025.106865

Authors: Wenjuan Li, Xinghua Liu, Junqi Su, Tianxiang Cui

Institutions: Shandong University of Finance and Economics, City University of Hong Kong, University of Nottingham Ningbo China

Language: English

Topic: Financial Distress and Bankruptcy Prediction

Subfield: Accounting

Field: Business, Management and Accounting

Domain: Social Sciences

Open Access status: hybrid

Estimated APC paid: $3,730

Funders: Ningbo Municipal People's Government, Natural Science Foundation of Ningbo, University of Nottingham Ningbo China, Natural Science Foundation of Zhejiang Province