Artificial Intelligence for Risk Mitigation in the Financial Industry

7The Extraction of Features That Characterize Financial Fraud Behavior by Machine Learning Algorithms

George X. Yuan^1,2,3, Yuanlei Luo^4‡, Lan Di⁵†, Yunpeng Zhou³, Wen Chen³, Yiming Liu³ and Yudi Gu⁶*

¹College of Science, Chongqing University of Technology, Chongqing, China

²Business School, Chengdu University, Chengdu, China

³Shanghai Hammer Digital Tech. Co. Ltd. (Hammer), Shanghai, China

⁴Research Center, China Institute of Ocean Engineering, Beijing, China

⁵School of AI and Computer Science, Jiangnan Univ., Wuxi, China

⁶Center of Information Construct and Management, Jiangnan University, Wuxi, China

Abstract

The purpose of this paper is to discuss how to use machine learning algorithms to screen the features and related applications of the characteristic indicators in describing a company’s financial fraud behavior under the framework of big data analysis. In particular, we first screen the “characteristic indicator” (features) related to the financial anomalies due to a company’s fraudulent financial reports then extract (fraud) features based on the corresponding fraudulent actions. In addition, the validation test is carried out to test the ability of fraud features’ (indicators’) performance in identifying and providing warning signals on fraud events. Specifically, by extracting the characteristics (features) of fraudulent financial behaviors related to financial frauds, based on traditional (structure) financial data and unstructured corporate governance structures, ...