Advancing financial risk management: A transparent framework for effective fraud detection
A transparent hybrid model integrating XGBoost and Genetic Programming enables interpretable, effective financial fraud detection outperforming traditional techniques.
What it examines
This study introduces a hybrid XGB-GP model that integrates XGBoost for feature selection and Genetic Programming for transparent fraud detection. It automates the selection of key financial ratios, including a novel Total Liabilities to Operating Costs ratio, to accurately identify fraudulent activities and overcome limitations of traditional methods.
What it concludes
Financial fraud and risk are closely linked. The XGB-GP model yields an interpretable equation from key ratios, offering a transparent, efficient tool for dynamic fraud monitoring. Its applications span auditing, regulatory compliance, and risk management, suggesting future research into integrating multi-source data for enhanced detection.
Evidence objects
Researchers introduce the XGB-GP hybrid model combining Extreme Gradient Boosting for feature selection with Genetic Programming for an interpretable fraud equation, overcoming machine learning black box issues through statistical filtering.
key_findings bullet 1 · key_findings · validation V0
Researchers reveal the unexpected power of the total liabilities-to-operating costs ratio; this metric, when combined with return on equity, current ratio, operating cash flow, and equity multiplier, predicts financial misreporting.
key_findings bullet 2 · key_findings · validation V0
Leveraging CSMAR database data with rigorous preprocessing, including outlier removal and oversampling, the study presents a transparent fraud detection framework advancing risk management, although limited non-financial indicators may reduce generalizability.
key_findings bullet 3 · key_findings · validation V0
This paper presents a transparent framework for financial fraud detection employing genetic programming to generate explicit expressions. Its originality is clear as evolutionary methods fuse with established risk indicators, providing a novel, interpretable alternative to black-box models. Readers gain profound insight into advanced quantitative risk management through progressive model evolution.
key_findings bullet 4 · key_findings · validation V0
Raw abstract and provenance
- … financial characteristics and fraudulent activities, this study integrates both traditional financial indicators and all generated financial … financial and non-financial ratios using …
Source row: 121 · abstract type: snippet